Introduction to Algorithms October 13, 2011

Massachusetts Institute of Technology 6.006 Fall 2011

Professors Erik Demaine and Srini Devadas Quiz 1 Solutions

Quiz 1 Solutions
Problem 1. [2 points] Write your name on top of each page.

Problem 2. Asymptotics & Recurrences [20 points] (3 parts)

(a) [10 points] Rank the following functions by increasing order of growth. That is, find
any arrangement g1 , g2 , g3 , g4 , g5 , g6 , g7 , g8 of the functions satisfying g1 = O(g2 ),
g2 = O(g3 ), g3 = O(g4 ), g4 = O(g5 ), g5 = O(g6 ), g6 = O(g7 ), g7 = O(g8 ).

  
π n n √
f1 (n) = n f2 (n) = π f3 (n) = f4 (n) = 2 n
5
   
n 4 2 4 n
f5 (n) = f6 (n) = 2log n
f7 (n) = n5(log n) f8 (n) = n
n−4 4

Solution: f1 (n), f5 (n), f3 (n), f8 (n), f7 (n), f6 (n), f4 (n), f2 (n)

Scoring: We computed the score for this problem as ROUND(10 · NL−1 −1

), where N is
the number of functions (N = 8 for this instance) and L is the length of the longest
common subsequence between our solution and the student’s answer.
The intuition behind the longest common subsequence is that we want to cross out as
few functions as possible from a student’s answer, such that the remaining functions
will be correctly ordered. Who said the 6.006 staff isn’t nice?
We used NL−1−1
to normalize the scores, because a completely wrong answer will still
share a common subsequence of length 1 with the correct answer.
The longest common subsequence can be computed using Dynamic Programming,
which will be taught in 6.006 towards the end of the term.

(b) [5 points] Find a solution to the recurrence T (n) = T ( n3 ) + T ( 2n

3
) + Θ(n).

Solution: Draw recursion tree. At each level, do Θ(n) work. Number of levels is
log3/2 n = Θ(lg n), so guess T (n) = Θ(n lg n) and use the substitution method to
verify guess.
6.006 Quiz 1 Solutions Name 2

(c) [5 points] Find an asymptotic solution of the following recurrence. Express your
answer using Θ-notation, and give a brief justification.
√ 
T (n) = log n + T n

Solution: T (n) = Θ(log n).

To see this, note that if we expand out T (n) by continually replacing T (n) with its
formula, we get:
 
√ √ √
T (n) = log n + log n + log n + log n + ...

√ √
= log n + 12 log n + 21 log n + 12 log n + ...
= log n + 12 log n + 41 log n + 81 log n + . . .
= Θ(log n)
6.006 Quiz 1 Solutions Name 3

Problem 3. True/False [18 points] (9 parts)

Circle (T)rue or (F)alse. You don’t need to justify your choice.

(a) T F [2 points] Binary insertion sorting (insertion sort that uses binary search to find
each insertion point) requires O(n log n) total operations.

Solution: False. While binary insertion sorting improves the time it takes to
find the right position for the next element being inserted, it may still take O(n)
time to perform the swaps necessary to shift it into place. This results in an O(n2 )
running time, the same as that of insertion sort.

(b) T F [2 points] In the merge-sort execution tree, roughly the same amount of work is
done at each level of the tree.

Solution: True. At the top level, roughly n work is done to merge all n ele-
ments. At the next level, there are two branches, each doing roughly n/2 work to
merge n/2 elements. In total, roughly n work is done on that level. This pattern
continues on through to the leaves, where a constant amount of work is done on
n leaves, resulting in roughly n work being done on the leaf level, as well.

(c) T F [2 points] In a BST, we can find the next smallest element to a given element in
O(1) time.

Solution: False. Finding the next smallest element, the predecessor, may re-
quire traveling down the height of the tree, making the running time O(h).

(d) T F [2 points] In an AVL tree, during the insert operation there are at most two
rotations needed.

Solution: True. The AVL property is restored on every operation. Therefore,

inserting another item will require at most two rotations to restore the balance.

(e) T F [2 points] Counting sort is a stable, in-place sorting algorithm.

Solution: False. Counting sort is stable. It is not in-place, however, since we

must make additional space to store the counts of the various elements. This
space requirement grows as the size of the input increases. Additionally, we have
to make a separate output array to produce the answer using counting sort.

(f) T F [2 points] In a min-heap, the next largest element of any element can be found
in O(log n) time.
6.006 Quiz 1 Solutions Name 4

Solution: False. A min-heap cannot provide the next largest element in O(log n)
time. To find the next largest element, we need to do a linear, O(n), search
through the heap’s array.

(g) T F [2 points] The multiplication method satisfies the simple uniform hashing as-
sumption.

Solution: False. We don’t really know of hash functions that satisfy the simple
uniform hashing assumption.

(h) T F [2 points] Double hashing satisfies the uniform hashing assumption.

Solution: False. The notes state that double hashing ‘comes close.’ Double
hashing only provides n2 permutations, not n!.

(i) T F [2 points] Python generators can be used to iterate over potentially infinite count-
able sets with O(1) memory.

Solution: True. Python generators do not require the whole set to reside in
memory to iterate over it, making this assertion true.
6.006 Quiz 1 Solutions Name 5

Problem 4. Peak Finding (again!) [20 points] (2 parts)

When Alyssa P. Hacker did the first 6.006 problem set this semester, she didn’t particularly like
any of the 2-D peak-finding algorithms. A peak is defined as any location that has a value at least
as large as all four of its neighbors.
Alyssa is excited about the following algorithm:

1. Examine all of the values in the first, middle, and last columns of the matrix to find the
maximum location .
2. If  is a peak within the current subproblem, return it. Otherwise, it must have a neighbor p
that is strictly greater.
3. If p lies to the left of the central column, restrict the problem matrix to the left half of the ma-
trix, including the first and middle columns. If p lies to the right of the central column, restrict
the problem matrix to the right half of the matrix, including the middle and last columns.
4. Repeat steps 1 through 3 looking at the first, middle, and last rows.
5. Repeat steps 1 through 4 until a peak is found.

Consider the 5 × 5 example depicted below. On this example, the algorithm initially examines the
first, third, and fifth columns, and finds the maximum in all three. In this case, the maximum is the
number 4. The number 4 is not a peak, due to its neighbor 5.

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 5 0 0 0 4 5 0 0 0 4 5 0 0 0
0 0 1 2 0 0 0 1 2 0 0 0 1 2 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 6 0 7 0 0 6 0 7 0 0 6 0 7 0

The number 5 is to the left of the middle column, so we restrict our view to just the left half of the
matrix. (Note that we include both the first and middle columns.) Because we examined columns
in the previous step, we now examine the first, middle, and last rows of the submatrix. The largest
value still visible in those rows is 6, which is a peak within the subproblem. Hence, the algorithm
will find the peak 6.

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 5 0 0 0 4 5 0 0 0 4 5 0 0 0
0 0 1 2 0 0 0 1 2 0 0 0 1 2 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 6 0 7 0 0 6 0 7 0 0 6 0 7 0
6.006 Quiz 1 Solutions Name 6

(a) [5 points] What is the worst-case runtime of Alyssa’s algorithm on an m × n matrix

(m rows, n columns), in big-Θ notation? Give a brief justification for your answer.

Solution: Let S(m, n) be the runtime of the algorithm when run on an m × n matrix
starting with columns. Let T (m, n) be the runtime of the algorithm when run on
an m × n matrix starting with rows. Then S(m, n) ≤ T (m, n/2 + 1) + Θ(m) and
T (m, n) ≤ S(m/2+1, n)+Θ(n). Hence, S(m, n) ≤ Θ(m+n)+S(m/2+1, n/2+1).
When we resolve this recurrence relation, we get S(m, n) = O(m + n). In the case
of a square n × n matrix, we get an asymptotic runtime of Θ(n).

(b) [15 points] Does Alyssa’s algorithm return a peak in all cases? If so, give a short
proof of correctness. Otherwise, provide a counterexample for the algorithm.

Solution: The following is an example of a matrix where the algorithm will return
the wrong value:

0 0 0 0 0
4 5 0 0 0
0 0 1 2 0
0 0 0 0 0
0 0 0 7 0
6.006 Quiz 1 Solutions Name 7

Problem 5. Who Let The Zombies Out? [20 points] (2 parts)

In an attempt to take over Earth, evil aliens have contaminated certain water supplies with a virus
that transforms humans into flesh-craving zombies. To track down the aliens, the Center for Dis-
ease Control needs to determine the epicenters of the outbreak—which water supplies have been
contaminated. There are n potentially infected cities C = {c1 , c2 , . . . , cn }, but the FBI is certain
that only k cities have contaminated water supplies.
Unfortunately, the only known test to determine the contamination of a city’s water supply is to
serve some of that water to a human and see whether they turn ravenous. Several brave volunteers
have offered to undergo such an experiment, but they are only willing to try their luck once. Each
volunteer is willing to drink a single glass of water that mixes together samples of water from any
subset C  ⊆ C of the n cities, which reveals whether at least one city in C  had contaminated
water.
Your goal is to use the fewest possible experiments (volunteers) in order to determine, for each city
ci , whether its water was contaminated, under the assumption that exactly k cities have contami-
nated water. You can design each experiment based on the results of all preceding experiments.

(a) [10 points] You observe that, as in the comparison model, any algorithm can be
viewed as a decision tree where a node corresponds to an experiment with two out-
comes (contaminated or not) and thus two children. Prove a lower bound of Ω(k lg nk )
on the number of experiments that must be done to save the world. Assume that
lg x! ∼ x lg x and that lg(n − k) ∼ lg n (which is reasonable when k < 0.99n).

Solution:
n The number of possible outcomes—whichn cities are contaminated—is
k
. Thus any decision tree must have at least k leaves. Because a decision tree
is binary, it must therefore have height at least
 
n n!
lg = lg ,
k k!(n − k)!

which by the first assumption is

∼ n lg n − k lg k − (n − k) lg(n − k) = n[lg n − lg(n − k)] + k[lg(n − k) − lg k],

which by the second assumption is

n
∼ k[lg n − lg k] = k lg ,
k
which is our desired lower bound.
6.006 Quiz 1 Solutions Name 8

(b) [10 points] Save the world by designing an algorithm to determine which k of the n
cities have contaminated water supplies using O(k lg n) experiments. Describe and
analyze your algorithm.

Solution: The algorithm is based on divide and conquer: divide the n cities into two
groups of size n/2; test each group for contamination (using two experiments); and
recurse into each contaminated group. The recursion tree has exactly k leaves, and the
height of the tree is at most lg n, so the number of internal nodes leading to the leaves
is at most k lg n. Each internal node costs 2, for a total cost of O(k lg n).
In fact, it is possible to prove an O(k lg nk ) bound on the same algorithm. To minimize
the number of shared nodes among the k paths from root to leaves, the worst case is
when the recursion tree branches for the first lg k levels (to get enough leaves), and
then has k straight paths for the number of levels: lg n − lg k = lg nk . There are O(k)
nodes in the top branching, and O(k lg nk ) nodes in the bottom paths.
6.006 Quiz 1 Solutions Name 9

Problem 6. Shopping Madness [20 points] (3 parts)

Ben Bitdiddle was peer-pressured into signing up for the tryouts in a shopping reality TV show,
and he needs your help to make it past the first round. In order to qualify, Ben must browse a
store’s inventory, which has N items with different positive prices P [1], P [2], . . . , P [N ], and the
challenge is to spend exactly S dollars on exactly K items, where K is a small even integer.
In your solutions below, you may use a subroutine S UBSETS(k, T) which iterates over all the k-
element subsets of a set T, in time O(k · |T|k ), using O(k) total space. Note that if your code holds
onto the results of S UBSETS, it may end up using more than O(k) space.

(a) [5 points] Write pseudo-code for a data structure that supports the following two
operations.
I NIT(N, K, P ) — preprocesses the P [1 . . . N ] array of prices, in O(K ·N K ) expected
time, using O(K · N K ) space, to be able to answer the query below.
BAG(S) — in O(1) expected time, determines whether K of the items have prices
summing to S, and if so, returns K indices b1 , b2 , . . . , bK such that S = K
i=1 P [bi ].

Solution:

I NIT(N, K, p)
1 h ← empty hash table
2 for c ← S UBSETS(K, {1 . . . N })
3 do s ← K i=1 Pci
4 h[s] ← c

BAG(S)
1 if S ∈ h
2 then return h[S]
3 else return N IL
6.006 Quiz 1 Solutions Name 10

(b) [10 points] Write pseudo-code for a function P WN -C ONTEST(N, S, K, P ) that deter-
mines whether K of the items have prices summing to S, and if so, returns K indices
b1 , b2 , . . . , bK such that S = K
i=1 P [bi ]. Unlike part (a), P WN -C ONTEST should run
in O(K · N ) and use O(K · N K/2 ) space.
K/2

Solution:

P WN -C ONTEST(N, S, K, p)
1 h ← empty hash table
2 for c ← S UBSETS(K/2, {1 . . . N })
K/2
3 do s ← i=1 P ci
4 h[s] ← c
5 for c ← S UBSETS(K/2, {1 . . . N })
6 do s ← S − K/2 i=1 Pci
7 if s ∈ h
8 then return c + h[s]
9 return N IL
6.006 Quiz 1 Solutions Name 11

(c) [5 points] Analyze the running time of your pseudo-code for the previous part.

Solution: The following table shows a line-by-line analysis of our pseudo-code.

Line Time Number of iterations Total time
1 O(1) 1 O(1)
K/2
2 K ·N 1 K · N K/2
3 O(K) N K/2 K · N K/2
K/2
4 O(K) N K · N K/2
5 K · N K/2 1 K · N K/2
6 O(K) N K/2 K · N K/2
K/2
7 O(1) N N K/2
8 O(K) 1 O(K)
9 O(1) 1 O(1)
The total running time is the maximum in the “Total time” column, which is K ·N K/2 ,
as requested.
6.006 Quiz 1 Solutions Name 12

Problem 7. When I Was Your Age. . . [20 points] (2 parts)

In order to design a new joke for your standup comedy routine, you’ve collected n distinct mea-
surements into an array A[1 . . . n], where A[i] represents a measurement at time i. Your goal is to
find the longest timespan i . . . j, i.e., maximize j − i, such that A[i] < A[j]. 1 Note that the values
in between A[i] and A[j] do not matter. As an example, consider the following array A[1 . . . 7]:

A[1] = 14 A[2] = 6 A[3] = 8 A[4] = 1 A[5] = 12 A[6] = 7 A[7] = 5

Your algorithm should return a span of 4 since A[2] = 6 and A[6] = 7. The next biggest span is
A[4] = 1 to A[7] = 5.

(a) [5 points] Give an O(n)-time algorithm to compute the minimums of the prefix
A[1 . . . k] for each k, and store in MA[k]: MA[k] = minki=1 A[i].

Solution: MA[i] can be computed incrementally. Initially, MA[1] = A[1]. MA[j] =

min(A[j], MA[j − 1]). This takes O(n) time.

(b) [15 points] Using the MA[i] computed above, give an O(n log n)-time algorithm to
maximize j − i subject to A[i] < A[j].
Hint: The M A is a sorted array.

Solution: Consider a single element A[j]. If we have M A[1 . . . j − 1] we want

to find an index i such that M A[i] < A[j] but M A[i − 1] ≥ A[j]. This implies that
M A[i] = A[i] is the unique minimum element of A[1 . . . i]. This gives us an A[i], A[j]
pair and we can compute j − i for this pair.
We do a binary search over the indices [1, j−1]. We start with j/2, and we test whether
M A[j/2] < A[j] or not. If M A[j/2] is less than A[j], we recurse on M A[1 . . . (j/2 −
1)]. If M A[j/2] is larger than A[j], we recurse on M A[j/2 + 1 . . . j − 1]. We halve
the number of possible indices for i each time, until we find the right i for this j. This
takes O(log n) time. We do this for each j – hence the O(n log n). After we have the
right i, j pairs, we pick the one that maximizes j − i.

1
The joke could be along these lines: “You thought time j was bad with A[j]? Back in time i, we only had A[i]!”
MIT OpenCourseWare
http://ocw.mit.edu

6.006 Introduction to Algorithms

Fall 2011

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.