CS 4310 HOMEWORK SET 1

PAUL MILLER

Section 2.1 Exercises Exercise 2.1-1. Using Figure 2.2 as a model, illustrate the operation of INSERTION-SORT on the array A = 31, 41, 59, 26, 41, 58 . Solution: Assume A is indexed from 1. Begin with A = [31, 41, 59, 26, 41, 58]. Since A[2] is already in the right spot relative to A[1], we do nothing when j = 2. Similarly, since A[3] is already in the right spot relative to A[1]...A[2], we do nothing again when j = 3. For j = 4, we nd that A[4] == 26 is out of place, so we move it to its proper location in front of 31 and shift the other elements over to the right, giving A = [26, 31, 41, 59, 41, 58]. For j = 5, we nd that A[5] == 41 is out of place, so we move it to its proper place in front of the previous instance of 41 and shift elements over to accommodate it, giving A = [26, 31, 41, 41, 59, 58]. Finally, for j = 6, we nd A[6] == 58 is out of place, so we put it in its place and shift elements over to accommodate it, giving us nally A = [26, 31, 41, 41, 58, 59], the sorted sequence.

Exercise 2.1-3. Consider the searching problem: Input: A sequence of n numbers A = a1 , a2 , . . . , an and a value v . Output: An index i such that v = A[i] or the special value NIL if v does not appear in A. Write pseudocode for linear search, which scans through the sequence, looking for v . Using a loop invariant, prove that your algorithm is correct. Make sure that your loop invariant fullls the three necessary properties. Solution: The following Python code implements linear search:
Date : October 1, 2009.
1

PAUL MILLER

1 2 3 4 5

def linear_search (A , v ): for i in range ( len (A) ): if A [i] == v: return i return NIL

(Note that range (len (A)) returns a list of integers from 0 to len (A)-1, inclusive. Also, Python arrays are indexed from 0, so an array of length n has indices 0 . . . n 1.) To show correctness, we shall use the following loop invariant: v {A[0], A[1], . . . , A[i 1]}. Initialization: At the beginning of the rst iteration, we have i = 0, so the statement is empty. Maintenance: Suppose the invariant is not maintained after the k -th iteration. We would then have some i k 1 such that A[i] = v . However, in that case, the code would have returned the value i already, so there would not actually be a k -th iteration. This contradiction shows our initial assumption was incorrect, so the invariant is, in fact, maintained by the loop. Termination: Since the loop is a for loop over a nite sequence 0 ...len (A)-1, the loop will always terminate. If the algorithm nds v in the array A, we have A[i] = v , so the loop invariant is true (and the algorithm returns the index i). Otherwise, the loop terminates after len (A) iterations, in which case the invariant states that v {A[0], . . . , A[len(A) 1]}, which is the whole array A, so the invariant is again true, and the algorithm returns NIL. Therefore, the function linear search is correct by the loop invariant.

Section 2.2 Exercises Exercise 2.2-1. Express the function n3 /1000 100n2 100n + 3 in terms of -notation. Solution: Note: This is a badly worded problem. As written, I can write n3 /1000 100n2 100n + 3 = (n3 /1000 100n2 100n + 3). A

CS 4310 HOMEWORK SET 1

better rephrasing would be to nd all values of c for which n3 /1000 100n2 100n + 3 = (nc ). Since 1 if c = 3, n3 /1000 100n2 100n + 3 lim = 0 if c > 3, and n nc if c < 3, we have that n3 /1000100n2 100n+3 = (nc ) if and only if c = 3.

Exercise 2.2-2. Consider sorting n numbers stored in array A by rst nding the smallest element of A and exchanging it with the element in A[1]. Then nd the second smallest element of A, and exchange it with A[2]. Continue in this manner for the rst n 1 elements of A. Write pseudocode for this algorithm, which is known as selection sort. What loop invariant does this algorithm maintain? Why does it need to run for only the rst n 1 elements, rather than for all n elements? Give the best-case and worst-case running times of selection sort in -notation. Solution: The following pseudocode implements selection sort:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

def index_of_min (A ): smallest = 0 i = 0 while i < len ( A): if A [i] < A [ smallest ]: smallest = i i += 1 return smallest def selection_sort (A): i = 0 while i < len ( A) - 1: # A [i :] indicates A[i ] , A [i+1 ] , ... j = index_of_min (A [i: ]) + i # Swaps A[ i] and A [j] A[i] , A[ j] = A[ j] , A[i ] i += 1 return A

This algorithm maintains the loop invariant that after i iterations, the subarray consisting of the rst i elements of the array A is sorted.

PAUL MILLER

Indeed, after 1 iteration, the rst 1 elements of the array are sorted, since we have swapped the previous value of A[0] with min (A) (if necessary). The invariant is maintained because on the i-th iteration, we swap A[i-1] with min (A[i-1:]) (if necessary). It only needs to run for the rst n 1 elements of A because after the rst n 1 elements of A are sorted, the last element is guaranteed to be in the correct position. (If it were not, this would imply that the rst n 1 elements were not in sorted order, contradicting the loop invariant.) This then shows that the array is sorted upon termination of the loop. Since we must examine every element of an unsorted array to nd the index of the minimum value, the function index of min must always examine all n elements of an n-element array. Thus, selection sort always uses
n

n +n 1+n 2++1 =
j =1

j = n(n + 1)/2

comparisons to sort the array. Thus, selection sort uses (n2 ) comparisons in both the best and worst cases.

Section 2.3 Exercises Exercise 2.3-1. Using Figure 2.4 as a model, illustrate the operation of merge sort on the array A = 3, 41, 52, 26, 38, 57, 9, 49 . Solution: At the bottom level, we have eight subarrays of length 1: [3], [41], [52], [26], [38], [57], [9], [49]. These are then merged to form four sorted subarrays of length 2: [3, 41], [26, 52], [38, 57], [9, 49]. (Note: when merging, we always merge consecutive pairs of smaller subarrays according to Figure 2-4. In reality, it does not matter which pairs of sorted subarrays we merge this is simply the order in which the algorithm does things as presented in the text.) These then get merged to form two subarrays of length 4: [3, 26, 41, 52], [9, 38, 49, 57]. Finally, these are merged to form a single subarray of length 8: [3, 9, 26, 38, 41, 49, 52, 57],

CS 4310 HOMEWORK SET 1

and we are done.

Exercise 2.3-3. Use mathematical induction to show that when n is an exact power of 2, the solution of the recurrence T (n) = is T (n) = n lg n. Solution: Since T (2) = 2 lg 2 = 2 = 21 by substitution, the base case holds. For the induction hypothesis, assume that T (2k ) = 2k lg 2k . By the denition of our recurrence, for n = 2k+1 , we then have that T (2k+1 ) = 2T (2k ) + 2k+1 . By the induction hypothesis, T (2k ) = 2k lg 2k , so T (2k+1 ) = 2(2k lg 2k ) + 2k+1 = 2k+1 (lg 2k + 1) = 2k+1 lg 2k+1 , which is what we were after, so the result follows by induction. 2 2T (n/2) + n if n = 2, if n = 2k for k > 1

Exercise 2.3-5. Referring back to the searching problem (see Exercise 2.1-3), observe that if the sequence A is sorted, we can check the midpoint of the sequence against v and eliminate half of the sequence from further consideration. Binary search is an algorithm that repeats this procedure, halving the size of the remaining portion of the sequence each time. Write pseudocode, either iterative or recursive, for binary search. Argue that the worst-case running time of binary search is (lg n). Solution: The following pseudocode implements a recursive binary search, returning the index of the element v if it is present, or the value None if it is not found:
1 2 3 4 5

def binary_search(A , v , low , high ) : if low > high : return None mid = ( low + high ) / 2 if A [ mid ] < v:

6
6 7 8 9 10

PAUL MILLER

return binary_search (A , v , mid + 1 , high ) elif A[ mid ] > v : return binary_search (A , v , low , mid - 1 ) else : return mid

Since the algorithm eliminates at least 1/2 of the possible indices under consideration at each step, after at most lg n steps, we will have either found the value v or determined that it is not in the array. Thus, the algorithm terminates after at most (lg n) steps.

Exercise Problem 2-3. The following code fragment implements Horners rule for evaluating a polynomial
n

P (x) =
k =0

a k xk

= a0 + x(a1 + x(a2 + + x(an1 + xan ) )), given the coecients a0 , a1 , . . . , an and a value for x: 1y0 2in 3 while i 0 4 do y ai + x y 5 ii1 1. What is the asymptotic running time of this code fragment for Horners rule? 2. Write pseudocode to implement the naive polynomial-evaluation algorithm that computes each term of the polynomial from scratch. What is the running time of this algorithm? How does it compare to Horners rule? 3. Prove that the following is a loop invariant for the while loop in lines 3-5. At the start of each iteration of the while loop of lines 3-5,
n(i+1)

y=
k =0

ak+i+1 xk

CS 4310 HOMEWORK SET 1

Interpret a summation with no terms as equaling 0. Your proof should follow the structure of the loop invariant proof presented in this k chapter and should show that, at termination, y = n k =0 ak x . 4. Conclude by arguing that the given code fragment correctly evaluates a polynomial characterized by the coecients a0 , a1 , . . . , an . Solution: 1. This algorithm does precisely two additions and one multiplication per iteration of the while loop. Since the while loop terminates after n iterations, it follows that Horners rule computes P (x) using 2n additions and n multiplications, hence (n) time. 2. The following pseudocode implements the na ve polynomial evaluation algorithm. The sequence a is assumed to contain the coecients of P (x).
1 2 3 4 5 6 7 8

def naive_poly (a , x): k = len (a) if k == 1: return a [0 ] j = 1 for i in range (k- 1): j *= x return a [k- 1] * j + naive_poly (a[ 0:k -1] , x)

For a polynomial of degree n, this algorithm always does precisely j multiplications to compute the aj xj term, and precisely n 1 additions in total. Thus, the algorithm does 1 + 2 + + n 1 + n = n(n + 1)/2 multiplications and n 1 additions, implying this algorithm uses (n2 ) operations. 3. At the start of the rst iteration, we have i = n and y = 0, so
n(i+1) 1

ak+i+1 x =
k =0 k =0

ak+n+1 xk = 0

because the sum is empty, so the loop invariant holds. Suppose i = j with 0 j < k at the beginning of an iteration. By our loop invariant, we have
n(j +1)

y=
k =0

ak+j +1 xk

at the beginning of the iteration. The algorithm then sets y to

aj + x
k =0

ak+j +1 x = aj x +
k =0

ak+j +1 xk+1 .

PAUL MILLER

Next, the algorithm decrements i, so we now have j = i + 1. Thus, the previous expression becomes
n(i+1+1) n(i+2)

ai+1 x +
k =0

ak+(i+1)+1 x

k +1

= ai+1 x +
k =0 n(i+1)

ak+i+2 xk+1 am+i+1 xm

= ai+1 x0 +
n(i+1)

=
m=0

am+i+1 xm ,

so the invariant is maintained. Finally, at termination, we have i = 1, so the value of y is

ak+1+1 x =
k =0 k =0

a k xk ,

n k 4. Since the nal value of y is k =0 ak x , we conclude that this algorithm correctly evaluates the polynomial P (x) = a0 + a1 x + + an1 xn1 + an xn .

Big O problems1 For each pair of functions f (n), g (n), determine which of the following apply: (a) f (n) = (g (n)), (b) f (n) = O (g (n)) but f (n) = (g (n)), or (c) f (n) = (g (n)) but f (n) = g (n). Exercise 1. f (n) = 100n + log n, g (n) = n + (log n)2 . Solution: Since 100 + 1/n 100n + log n (by LH opitals rule) = lim lim 2 n 1 + (2 log n)/n n n + (log n) = 100, by the theorem given in class (see, e.g., the text by Brainerd & Landweber), we have f (n) = (g (n)).
1Not

part of Homework Set 1; from the text by Udi Manber.

CS 4310 HOMEWORK SET 1

Exercise 2. f (n) = log(n), g (n) = log(n2 ). Solution: Since g (n) = 2 log(n) = 2f (n), we have f (n) = (g (n)).

Exercise 3. f (n) = n2 /(log n), g (n) = n(log n)2 . Solution: Since n2 /(log n) n = lim 2 n n(log n) n (log n)3 1 ( by LH opitals rule) = lim n 3(log n)2 /n n = lim n 3(log n)2 1 = lim n 6(log n)/n n = lim n 6 log n 1 = lim (by LH opitals rule yet again) n 6/n n = lim = , n 6 lim the theorem from class tells us that f (n) = (g (n)) but f (n) = (g (n)).

Exercise 4. f (n) = (log n)log n , g (n) = n/(log n). Solution: Let n = eu for some u. We then have f (n) = uu and g (n) = (log u)/u. Then, f (n) uu uu1 = = . g (n) (log u)/u log u

10

PAUL MILLER

Since uu1 increases faster than u, and u 1 lim = lim u log u u 1/u = lim u
u

= , we know limu f (n)/g (n) = limn f (n)/g (n) = . Hence, our theorem tells us that f (n) = (g (n)) but f (n) = (g (n)).

Exercise 5. f (n) = n1/2 , g (n) = (log n)5 . Solution: Substituting n = e2u , we get that eu f (n) = . g (n) 32u5 Applying LH opitals rule ve times sequentially, we nally get eu f (n) eu lim = lim = = lim , 5 u 32u u 32 5! n g (n) so our theorem tells us that f (n) = (g (n)) but f (n) = (g (n)).

Exercise 6. f (n) = n2n , g (n) = 3n . Solution: Since f (n)/g (n) = n(2/3)n , substituting n = log2/3 u, we get f (n) = u log2/3 u. g (n) We thus conclude that f (n) = lim u log2/3 u = , n g (n) u therefore, we have f (n) = (g (n)) but f (n) = (g (n)). lim