Introduction to Algorithms

Massachusetts Institute of Technology 6.006 Spring 2020

Instructors: Erik Demaine, Jason Ku, and Justin Solomon Solution: Quiz 1

Solution: Quiz 1
• Do not open this quiz booklet until directed to do so. Read all the instructions on this page.
• When the quiz begins, write your name on the top of every page of this quiz booklet.
• You have 120 minutes to earn a maximum of 120 points. Do not spend too much time on
any one problem. Skim them all first, and attack them in the order that allows you to make
the most progress.
• You are allowed one double-sided letter-sized sheet with your own notes. No calculators,
cell phones, or other programmable or communication devices are permitted.
• Write your solutions in the space provided. Pages will be scanned and separated for grading.
If you need more space, write “Continued on S1” (or S2, S3) and continue your solution on
the referenced scratch page at the end of the exam.
• Do not spend time and paper rederiving facts that we have presented in lecture or recitation.
Simply cite them.
• When writing an algorithm, a clear description in English will suffice. Pseudo-code is not
required. Be sure to argue that your algorithm is correct, and analyze the asymptotic
running time of your algorithm. Even if your algorithm does not meet a requested bound,
you may receive partial credit for inefficient solutions that are correct.
• Pay close attention to the instructions for each problem. Depending on the problem,
partial credit may be awarded for incomplete answers.

Problem Parts Points

1: Information 2 2
2: Frequentest 2 8
3: Haphazard Heaps 2 10
4: Transforming Trees 2 10
5: Sorting Sock 4 20
6: Triple Sum 1 15
7: Where am i? 1 15
8: Methane Menace 1 20
9: Vapor Invite 1 20
Total 120

Name:

MIT Kerberos Username:

2 6.006 Solution: Quiz 1 Name

Problem 1. [2 points] Information (2 parts)

(a) [1 point] Write your name and email address on the cover page.
Solution: OK!

(b) [1 point] Write your name at the top of each page.

Solution: OK!
6.006 Solution: Quiz 1 Name 3

Problem 2. [8 points] Frequentest (2 parts)

The following two Python functions correctly solve the problem: given an array X of n positive
integers, where the maximum integer in X is k, return the integer that appears the most times in X.
Assume: a Python list is implemented using a dynamic array; a Python dict is implemented
using a hash table which randomly chooses hash functions from a universal hash family; and
max(X) returns the maximum integer in array X in worst-case O(|X|) time. For each function,
state its worst-case and expected running times in terms of n and k.

(a) [4 points] (b) [4 points]

def frequentest_a(X): def frequentest_b(X):

k = max(X) k = max(X)
H = {} A = []
for x in X: for i in range(k + 1):
H[x] = 0 A.append(0)
best = X[0] best = X[0]
for x in X: for x in X:
H[x] += 1 A[x] += 1
if H[x] > H[best]: if A[x] > A[best]:
best = x best = x
return best return best

i. Worst-case: i. Worst-case:
Solution: O(n2 ) Solution: O(n + k)

ii. Expected: ii. Expected:

Solution: O(n) Solution: O(n + k)

Common Mistakes: Saying doing a O(1)-time operation m times takes amortized O(m) time.
4 6.006 Solution: Quiz 1 Name

Problem 3. [10 points] Haphazard Heap (3 parts)

Array [A,B,C,D,E,F,G,H,I,J] represents a binary min-heap containing 10 items, where the
key of each item is a distinct integer. State which item(s) in the array could have the key with:

(a) the smallest integer

Solution: A

(b) the third smallest integer

Solution: B, C, D, E, F, G
Common Mistakes: Incorrectly assuming third smallest must be in either B or C.

(c) the largest integer

Solution: F, G, H, I, J
Common Mistakes: Thinking the largest integer must be in the bottom level
(instead of in any leaf)
6.006 Solution: Quiz 1 Name 5

Problem 4. [10 points] Transforming Trees (2 parts)

The tree below contains 8 items, where each stored item is an integer which is its own key.

93
85 38
23 63 11 13
17

(a) [6 points] Suppose the tree drawn above is the implicit tree of a binary max-heap H.
State the array representation of H, first before and then after performing the opera-
tion H.delete max().
Solution:
Before: [93, 85, 38, 23, 63, 11, 13, 17]
After: [85, 63, 38, 23, 17, 11, 13]
Common Mistakes:
• Not building heap correctly (e.g., confusing heap order with traversal order)
• Not correctly deleting min (by swapping root with last leaf and heapifying down)

(b) [4 points] Suppose instead that the original tree drawn above is a Sequence AVL
Tree S (note Sequence data structures are zero-indexed). The items in the leaves of
S in traversal order are (17, 63, 11, 13). Perform operation S.delete at(3) on S
including any rotations, and then list the items stored in the leaves of S in traversal
order, after the operation has completed. (You do not need to draw the tree.)
Solution: (17, 85, 11, 13)
Common Mistakes:
• Indexing by heap order instead of traversal order (or using 1-indexing)
• Listing full traversal order (rather than just the leaves)
• Deleting the item but not rebalancing the tree to satisfy the AVL Property
6 6.006 Solution: Quiz 1 Name

Problem 5. [20 points] Sorting Sock (4 parts)

At Wog Hearts School of Wizcraft and Witcherdry, n incoming students are sorted into four houses
by an ancient magical artifact called the Sorting Sock. The Sorting Sock first sorts the n students
by each of the four houses’ attributes and then uses the results to make its determinations. For each
of the following parts, state and justify what type of sort would be most efficient. (By “efficient”,
we mean that faster correct algorithms will receive more points than slower ones.)

(a) [5 points] For House Puffle Huff, students must be sorted by friend number, i.e.,
how many of the other n − 1 incoming students they are friends with, which can be
determined in O(1) time.
Solution: Friend numbers are non-negative integers less than n, so we can use count-
ing sort to sort the students in worst-case O(n) time. (Radix sort also works with the
same running time.) Since we have to compute friend number for each student, any
algorithm will take at least Ω(n) time, so this is optimal.
Common Mistakes: Using a sort that is not O(n)

(b) [5 points] For House Craven Law, students must be sorted by the weight of their
books. Book weights cannot be measured precisely, but the Sorting Sock has a scale
that can determine in O(1) time whether one set of books has total weight greater than,
less than, or equal to another set of books.
Solution: A scale weighing is a comparison with a constant number of outcomes, so
the comparison sort Ω(n log n) lower bound applies. So we cannot do better than by
using an worst-case O(n log n) sorting algorithm, e.g., merge sort, using the scale to
compare one student’s books against another’s.
Common Mistakes: Insufficient justification for why O(n log n) is optimal
6.006 Solution: Quiz 1 Name 7

(c) [5 points] For House Driven Gore, students must be sorted by bravery, which can’t
be directly measured or quantified, but for any set of students, the Sorting Sock can
determine the bravest among them in O(1) time, e.g., by presenting the students with
a scary situation.
Solution: We can’t quantify bravery, so we can’t hope to use any integer-based
algorithms. However, the Sorting Sock can find a student of maximum bravery in
O(1) time, so we repeatedly find and select a bravest student among all previously
unselected students in worst-case O(n) time, which is again optimal. (This is priority
queue sort, using the Sorting Sock as the priority queue to find the maximum.)
Common Mistakes: Arguing a Ω(n log n) lower bound

(d) [5 points] For House Leather Skin, students must be sorted by their magical lineage:
how many of a student’s ancestors within the previous 3dlog ne + 4 generations were
magical. Recall that humans, magical or not, always have two parents in the previous
generation, unlike binary tree nodes which have at most one. Assume the Sorting Sock
can compute the magical lineage of a student in O(1) time.
Solution: Each student has at most 2k ancestors in the kth generation preceding.
Thus
Pthe number of wizard ancestors will be a non-negative number bounded above
3dlog ne+4 k
by i=1 2 < 23(log n+1)+5 = 28 23 log n = 28 n3 log 2 = O(nc ) for any c > 3 log 2.
Thus we can use radix sort to sort the students by their magical lineage in worst-case
O(n) time, which is again optimal.
Common Mistakes:
• Claiming 3dlog ne + 4 ancestors instead of O(23dlog ne+4 )
• Saying that 23dlog ne+4 = O(n) and using counting sort
8 6.006 Solution: Quiz 1 Name

Problem 6. [15 points] Triple Sum

Given three arrays A, B, C, each containing n integers, give an O(n2 )-time algorithm to find
whether some a ∈ A, some b ∈ B, and some c ∈ C have zero sum, i.e., a + b + c = 0. State
whether your running time is worst-case, expected, and/or amortized.
Solution: We present both expected and worst-case solutions (both were accepted for full points).
Expected Time
For each pair of numbers (a, b) ∈ A × B, store a + b in a hash table H. Then return Yes if −c
appears in H for any c ∈ C, and return No otherwise.
Correctness: If any −c appears in H for any c ∈ C, then −c = a0 + b0 for some (a0 , b0 ) ∈ A × B
so a0 + b0 + c = 0. Otherwise, there is no c ∈ C for which −c = a0 + b0 for any (a0 , b0 ) ∈ A × B.
Running Time: There are O(n2 ) pairs in A × B, so inserting them into H takes expected O(n2 )
time. Then checking whether each −c appears in H takes expected O(1) time each, and expected
O(n) in total. So this algorithm runs in expected O(n2 ) time.
Continued on scratch paper S1 for worst-case solution...
Common Mistakes:

• Using counting/radix sort or creating a direct access array (no bound on u so not efficient)
• Saying n insertions into a hash table gives an amortized bound
• Checking all triples in Ω(n3 ) time
6.006 Solution: Quiz 1 Name 9

Problem 7. [15 points] Where Am i?

Given a Sequence AVL Tree T containing n nodes, and a pointer to a node v from T , describe an
O(log n)-time algorithm to return the (zero-indexed) index i of node v in the traversal order of T .
(Recall that every node u in a Sequence AVL Tree T stores an item u.item, parent u.parent,
left child u.left, right child u.right, subtree height u.height, and subtree size u.size.)
Solution: Our algorithm will be to walk up the tree from v to the root r of the Sequence AVL
Tree, counting the nodes preceding v in the traversal order along the way, since the number of
nodes preceding v in the tree is equivalent to v’s (zero-indexed) index.
Let #v (u) be the number of vertices preceding v in a vertex u’s subtree, where v is in the subtree
of u. Then #v (v) = v.left.size if v has a left child and zero otherwise; and can be computed
in O(1) time. Then, for every ancestor of u starting from v, we compute #v (u.parent) from
#v (u). There are two cases:

• Case 1, u is the left child of u.parent: then all the nodes preceding v in the subtree of
u.parent are in the subtree of u, so set #v (u.parent) = #v (u).
• Case 2, u is the right child of u.parent: then all nodes in the left subtree of u.parent
precede v (as does u), so set #v (u.parent) = 1 + u.parent.left.size + #v (u).

Then return #v (r), since this is the number of nodes preceding v in r’s subtree (i.e., the entire
tree). Correctness is argued within the algorithm description. This algorithm spends worst-case
O(1) work for each ancestor of v, so since the Sequence AVL Tree is balanced, the number of
ancestors is bounded by O(log n), and the algorithm runs in worst-case O(log n) time.
Common Mistakes:

• Finding v given i instead of finding i given v

• Breaking early (e.g., as soon as node is a left child, instead of continuing up tree)
• Walking down the tree from root (assuming the way to go to find v)
10 6.006 Solution: Quiz 1 Name

Problem 8. [20 points] Methane Menace

FearBird is a supervillian who has been making small holes in the methane gas pipe network of
mahtoG City. The network consists of n pipes, each labeled with a distinct positive integer. A
hole i is designated by a pair of positive integers (pi , di ), where pi denotes the label of the pipe
containing the hole, and di is a positive integer representing the distance of the hole from the front
of pipe pi . Assume any two holes in the same pipe pi will be at different distances from the front
of pi . When a new hole (pi , di ) is spotted, the city receives a report of the hole to keep track of.
The city will periodically patch holes using the following priority scheme:
• if each pipe contains at most one hole, patch any hole (if one exists);
• otherwise, among pairs of holes (pi , di ) and (pj , dj ) appearing on the same pipe, i.e., pi = pj ,
identify any pair with smallest distance |di − dj | between them, and patch one of them.
Describe a database supporting the following operations, where k is the number of recorded but
unpatched holes in the network at the time of the operation. State whether your running times are
worst-case, expected, and/or amortized.
initialize(H) Initialize the database with n holes H = {(p0 , d0 ), . . . , (pn−1 , dn−1 )},
with one hole on each pipe, in O(n) time
report(pi ,di ) Record existence of a hole in pipe pi at distance di in O(log k) time
patch() Patch any hole that follows the priority scheme above in O(log k) time
Solution: To implement the database, maintain the following data structures:
• A Set AVL tree Tp for each pipe p containing all the unpatched holes in p keyed by hole
distance
• A Hash Table D mapping each pipe p to its tree Tp
• A Binary Min Heap Q containing each consecutive pair of holes (p, d1 , d2 ) appearing on the
same pipe p with key being the distance |d2 − d1 | between them, and any lonely holes (p, d)
(holes that are alone on their pipes) with key ∞ (when multiple stored items have the same
key, we store them in a Hash Table keyed by (p, d1 , d2 ) or (p, d))
• A Hash Table C mapping each consecutive hole pair (p, d1 , d2 ) or lonely hole (p, d), to their
location in Q.
Some parenthetical notes on this solution:
• A Set AVL Tree can be used for C or Q to achieve identical bounds.
• A solution without augmentation was intended, so our solution does not use it. But it is also
possible to use augmentation:
– e.g., to combine C and Q into a single Set AVL Tree keyed the same as C but storing a
pointer to the min distance in subtree; or,
– e.g., on each Tp to maintain the min distance within the pipe (to compute this augmen-
tation efficiently, one would either need to maintain the distance to each hole’s succes-
sor/predecessor (not a subtree property), or augment by min/max distance in subtree to
be correct)
Operation descriptions: Continued on scratch paper S2...
6.006 Solution: Quiz 1 Name 11

Problem 9. [20 points] Vapor Invite

Vapor is an online gaming platform with n users. Each user has a unique positive integer ID di
and an updatable status, which can be either active or inactive. Every day, Vapor will post online
an active range: a pair of positive integers (a, b) with the property that every user having an ID
di contained in the range (i.e., with a ≤ di ≤ b) must be active. Vapor wants to post an active
range containing as many active users as possible, and invite them to play in a special tournament.
Describe a database supporting the following worst-case operations:
build(D) Initialize the database with user IDs D = {d0 , . . . , dn−1 },
setting all user statuses initially to active, in O(n log n) time
toggle status(di ) Toggle the status of the user with ID di ,
e.g., from active to inactive or vice versa, in O(log n) time
big active range() Return an active range (a, b) containing
the largest number of active users possible in O(1) time
Solution: To implement the database, maintain a single Set AVL Tree T containing each user ID
and their status, keyed by ID. In additional, augment each node x in T with four subtree properties:
• x.size: the number of IDs in the subtree (as discussed in lecture).
• x.suffix = (d, m): the smallest ID d in the subtree for which each of the m IDs d0 ≥ d in
the subtree is active, or None. Computable in O(1) time as either the suffix (dR , mR ) of the
right subtree, or if mR is equal to the size of the right subtree and x is active, return the suffix
(dL , mL ) of the left subtree but add mR + 1 to mL (or (x.key, mR + 1) if left suffix is None).
• x.prefix = (d, m): the largest ID d in the subtree for which each of the m IDs d0 ≤ d in
the subtree is active, or None. Computable in O(1) time as either the prefix (dL , mL ) of the
right subtree, or if mL is equal to the size of the left subtree and x is active, return the prefix
(dR , mR ) of the right subtree but add mL +1 to mR (or (x.key, mL +1) if right prefix is None).
• x.substring = (a, b, m): a, b are IDs from the subtree where each of the m IDs d in the
subtree with a ≤ d ≤ b is active and m is maximized. Computable in O(1) time by tak-
ing the max of the substring within either left or right subtree, or the substring spanning the
two subtrees if x is active. Specifically, consider the substrings of left and right subtrees,
(aL , bL , mL ) and (aR , bR , mR ) respectively, and then if x is active, consider the suffix of the
left subtree (dL , m0L ) and the prefix of the right subtree (dR , m0R ). Then depending on which
of (mL , m0L + 1 + m0R , mR ) is the largest, return (aL , bL , mL ), (dL , dR , m0L + 1 + m0R ), or
(aR , bR , mR ) respectively.
To implement build(D), build the Set AVL Tree T in worst-case O(n log n) time, maintaining
the custom augmentations during each insertion.
To implement toggle states(di), remove di from T in worst-case O(log n) time, toggle its
status in O(1) time, and then re-insert di into T in worst-case O(log n) time (again, maintaining
augmentations).
To implement big active range(), simply return the substring augmentation at the root in
worst-case O(1) time, which is correct by the definition of our augmentation.
Common Mistakes: See scratch paper S3...
12 6.006 Solution: Quiz 1 Name

SCRATCH PAPER 1. DO NOT REMOVE FROM THE EXAM.

You can use this paper to write a longer solution if you run out of space, but be sure to write
“Continued on S1” on the problem statement’s page.
Solution: (Problem 6 continued...)
Worst-case Solution
Sort A and B increasing using merge sort. We give a two-finger algorithm to determine whether
any a ∈ A and b ∈ B sum to a given −c. Doing this for each c ∈ C directly determines whether
a, b, c sum to 0.
Start with an index i = 0 into A and an index j = n − 1 into B and repeat the following procedure
until either i = n or j = −1:

• If A[i] + B[j] > −c, increase i.

• If A[i] + B[j] < −c, decrease j.
• Otherwise, A[i] + B[j] = −c, so return Yes.

If this procedure terminates without returning Yes, return No.

Correctness: We first prove the claim that the loop maintains the invariant that at the start of a
loop, that no A[i0 ] + B[j 0 ] = −c for any 0 ≤ i0 < i or any n − 1 ≥ j 0 > j. Proof by induction on
k = i − j. This invariant is trivially true at the start when k = i − j = −n + 1. Assume the claim
is true for all k < k ∗ = i∗ − j ∗ . If we are at the start of a loop with i = i∗ and j = j ∗ , we cannot
have yet returned, so at the start of the previous loop, there are two cases:

• Case 1, (i, j) = (i∗ − 1, j): i∗ − 1 − j = k ∗ − 1 < k ∗ , so by induction, no A[i0 ] + B[j 0 ] = −c

for any i0 < i∗ − 1 or j 0 > j ∗ . But we increased i when A[i∗ − 1] + B[j ∗ ] > −c, so since
B[j ∗ ] ≥ B[j 0 ] for all j 0 > j ∗ , then A[i∗ − 1] + B[j 0 ] > −c for all j 0 > j ∗ , restoring the
invariant.
• Case 2, (i, j) = (i∗ , j + 1): i∗ − (j + 1) = k ∗ − 1 < k ∗ , so by induction, no A[i0 ] + B[j 0 ] = −c
for any i0 < i∗ or j 0 > j ∗ + 1. But we decreased j when A[i∗ ] + B[j ∗ + 1] < −c, so since
A[i∗ ] ≤ A[i0 ] for all i0 < i∗ , then A[i0 ] + B[j 0 + 1] < −c for all i0 < i∗ , restoring the invariant.

So the claim is true. Then if the algorithm terminates without returning Yes, either i = n or
j = −1, so the claim implies that no integer from either A or B respectively can be in a triple with
c that sums to zero.
Running Time: Sorting A and B takes worst-case O(n log n) time. We perform the two-finger
algorithm n times. A single two-finger algorithm takes worst-case O(n): every loop does O(1)
work and either increases i or decreases j, so since the loop terminates when either i = n or
j = −1, the loop executes at most 2n = O(n) times. So the algorithm runs in worst-case O(n2 )
time in total.
6.006 Solution: Quiz 1 Name 13

SCRATCH PAPER 2. DO NOT REMOVE FROM THE EXAM.

You can use this paper to write a longer solution if you run out of space, but be sure to write
“Continued on S2” on the problem statement’s page.
Solution: (Problem 8 continued...)
To implement initialize(H), initialize empty D and C, and then for each (p, d) ∈ H, construct
an empty Set AVL Tree Tp , insert d into Tp , insert p into D mapping to Tp . Then build a hash table
D∞ on every (p, d) ∈ H, store it in Q with key ∞, and then insert each (p, d) into C mapping to
D∞ . Building D∞ takes expected O(n) time, and for each (p, d), this procedure takes O(1) time
(expected in the case of inserting into D). So it takes expected O(n) time in total, and maintains
the invariants of the database directly. (Note that we could not use a Set AVL Tree for D, as it
could take Ω(n log n) time to construct.)
To implement report(p, d), lookup p in D to find Tp , and then insert d into Tp .

• If d has no predecessor or successor in Tp , then insert (p, d) into Q with key ∞ and insert
(p, d) into C mapping to its location in Q.
• Otherwise d has a predecessor or successor in Tp .
– If d has a predecessor d1 and successor d2 , lookup (p, d1 , d2 ) in C to find it in Q, and
then remove (p, d1 , d2 ) from both C and Q.
– Otherwise, it has one of them d0 , so lookup (p, d0 ) in C to find it in Q and then remove
(p, d0 ) from both C and Q.
– In either case, if d has a predecessor d1 , add (p, d1 , d) to Q with key |d − d1 | and add
(p, d1 , d) to C pointing to its location in Q;
– and if d has a successor d2 , add (p, d, d2 ) to Q with key |d2 − d| and add (p, d, d2 ) to C
pointing to its location in Q.

This procedure does a constant number of worst-case O(log k) time or amortized expected O(1)
time operations, so this operation runs in amortized expected O(log k) time, and maintains the
invariants of the database by removing any consecutive pairs or lonely holes if they are no longer
consecutive or lonely, and adding any new consecutive pairs or lonely holes that may have been
introduced.
To implement patch(), delete the minimum item from Q containing one or two holes on pipe
p, remove it from C, lookup p in D to find Tp , and remove the relevant holes from Tp . This
procedure does a constant number of worst-case O(log k) time or amortized expected O(1) time
operations, so this operation runs in amortized expected O(log k) time, and maintains the invariants
of the database directly. It is correct because Q exactly implements the requested priority scheme:
consecutive pairs with smaller key have higher priority, and will only remove a lonely hole if there
are no consecutive pairs having finite key contained in Q.
Common Mistakes: See scratch page S3...
14 6.006 Solution: Quiz 1 Name

SCRATCH PAPER 3. DO NOT REMOVE FROM THE EXAM.

You can use this paper to write a longer solution if you run out of space, but be sure to write
“Continued on S3” on the problem statement’s page.
Common Mistakes: (For Problem 8)

• Taking a minimum in some pipe rather than a minimum over all pipes
• Not prioritizing pipes having more than one hole over those with only one
• Storing an AVL or heap on all pipes which cannot be maintained within the time bounds
• Attempting an augmentation-based solution incorrectly
• Claiming O(log n)-time insertion into a sorted array
• Initializing a direct access array of non-polynomially bounded size
• Using Sequence AVL Trees instead of Set AVL Trees

Common Mistakes: (For Problem 9)

• Assuming the max range always goes through the root

• Claiming a substring augmentation without showing how to maintain it in O(1) time
• Using 2 AVL Trees, one of active and one of inactive (doesn’t help to find largest range).
• Substring augmentation maintenance doesn’t consider substrings containing the root
MIT OpenCourseWare
https://ocw.mit.edu

6.006 Introduction to Algorithms

Spring 2020

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