CS 170, Fall 2024 Discussion 3 P. Raghavendra and S.

Garg

Note: Your TA probably will not cover all the problems. This is totally fine, the discussion worksheets
are deliberately made long so they can serve as a resource you can use to practice, reinforce, and build
upon concepts discussed in lecture, readings, and the homework.

1 Graph Traversal
A

B E C

I D

G H F J

(a) Recall that given a DFS tree, we can classify edges into one of four types:
• Tree edges are edges in the DFS tree,
• Back edges are edges (u, v) not in the DFS tree where v is the ancestor of u in the DFS
tree
• Forward edges are edges (u, v) not in the DFS tree where u is the ancestor of v in the
DFS tree
• Cross edges are edges (u, v) not in the DFS tree where u is not the ancestor of v, nor is v
the ancestor of u.
For the directed graph above, perform DFS starting from vertex A, breaking ties alphabetically.
As you go, label each node with its pre- and post-number, and mark each edge as Tree, Back,
Forward or Cross.
(b) A strongly connected component (SCC) is defined as a subset of vertices in which there exists
a path from each vertex to another vertex. What are the SCCs of the above graph?

(c) Collapse each SCC you found in part (b) into a meta-node, so that you end up with a graph of
the SCC meta-nodes. Draw this graph below, and describe its structure.

Solution:

This content is protected and may not be shared, uploaded, or distributed. 1 of ??

CS 170, Fall 2024 Discussion 3 P. Raghavendra and S. Garg

A61

T C
T

B23 E45 C720

C C B

C 16 11
I13 D10 T

T T T

B T T
G15
14
17
H12 F918 J819
(a)
(b)
{A}, {B}, {E}, {G, H, I}, {C, J, F, D}

CJF D GHI A E B

(c)

2 Graph Short Answer

For each of the following, either prove the statement is true or give a counterexample to show it is
false. Note that pre(v) and post(v) denote that pre-order and post-order values of v.

(a) If (u, v) is an edge in an undirected graph and during DFS, post(v) < post(u), then u is an
ancestor of v in the DFS tree.

(b) In a directed graph, if there is a path from u to v and pre(u) < pre(v) then u is an ancestor of
v in the DFS tree.

This content is protected and may not be shared, uploaded, or distributed. 2 of ??

CS 170, Fall 2024 Discussion 3 P. Raghavendra and S. Garg

(c) We can modify the SCC algorithm from lecture, so that, the DFS on G is done in decreasing
pre-order of G instead of decreasing post-order of GR . This modified algorithm is also correct.

(d) We can modify the SCC algorithm from lecture so that the first DFS is run on G and the second
DFS is run on GR . This modified algorithm is also correct.

Solution:

(a) True. There are two possible cases: pre(u) < pre(v) < post(v) < post(u) or pre(v) < post(v) <
pre(u) < post(u). In the first case, u is an ancestor of v. In the second case, v was popped off
the stack without looking at u. However, since there is an edge between them and we look at
all neighbors of v, this cannot happen.
(b) False. Consider the following case where we DFS starting from w:

w16
T

B
u32 v45

(c) False. Consider a 2-vertex graph with a single directed edge, i.e. G = (V = {v1 , v2 }, E =
{(v1 , v2 )}). Then if we start DFS from v2 , the vertex with the largest pre-order in G would be
v1 . However, we know that v1 is not the sink SCC and so this modified algorithm would return
the wrong answer.
(d) True. The SCCs of G are the same as the SCCs of GR .

This content is protected and may not be shared, uploaded, or distributed. 3 of ??

CS 170, Fall 2024 Discussion 3 P. Raghavendra and S. Garg

3 Finding Clusters
We are given a directed graph G = (V, E), where V = {1, . . . , n}, i.e. the vertices are integers in the
range 1 to n. For every vertex i we would like to compute the value m(i) defined as follows: m(i) is
the smallest j such from which you can reach vertex i. (As a convention, we assume that i is reachable
from i.)

(a) Show that the values m(1), . . . , m(n) can be computed in O(|V | + |E|) time. Please provide
an algorithm description and runtime analysis; proof of correctness is not required.

Solution: The algorithm is as follows.

procedure DFS-Clusters(G)
while there are unvisited nodes in G do
Run DFS on G starting from the numerically-first unvisited node j
for i visited by this DFS do m(i) := j
To see that this algorithm is correct, note that if a vertex i is assigned a value then that value
is the smallest of the nodes that can reach it in G, and every node is assigned a value because
the loop does not terminate until this happens.
The running time is O(|V | + |E|) since the algorithm is just a modification of DFS.
(b) Suppose we instead define m(i) to be the smallest j that can be reached from i, instead of the
smallest j from which you can reach i. How should you modify your answer to part (a) to work
in this case? Solution: We start by reversing all edges in G, i.e. replacing each edge (u, v) with
the edge (v, u), and then just using our algorithm from the previous part. This works because
in the reversed graph, we can reach j from i if and only if we can reach i from j in the original.

This content is protected and may not be shared, uploaded, or distributed. 4 of ??

CS 170, Fall 2024 Discussion 3 P. Raghavendra and S. Garg

4 Not So Exciting Scheduling

PNP University requires students to finish all prerequisites for a certain class before taking it; however,
they made some mistakes when assigning prerequisites. Thus, some classes at PNP University are
potentially NP (Not Possible to take) because it may be impossible to take all its prerequisite classes
whilst following proper prerequisite rules. For example, if CS 61A is a prereq of CS 61B, CS 61B is a
prereq of CS 170, and CS 170 is mistakenly listed as a prereq of CS 61A, then all three classes would
be NP.
Due to these mistakes, students wish to figure out whether their classes can all be taken or not while
following prerequisites. Their n classes are labelled with unique identifiers {c1 , c2 , . . . , cn }, and the
set of m prerequisites in the form [ci , cj ] indicate that ci must be taken before cj .
Design an algorithm that outputs a potential scheduling of classes (i.e. an ordering of classes to take)
if there are no NP classes; return false otherwise. Please provide an algorithm description and
a runtime analysis; proof of correctness is not required.
Solution:
Algorithm: Create a node for each class. For each prerequisite [ci , cj ], construct a directed edge
from cj to ci . Run DFS on the resulting graph while keeping track of pre and post numbers. (Cycle
detection) If a back edge exists, then output false. Otherwise sort the nodes on post numbers, from
highest to lowest (topological sort on DAG) or use a stack to keep track of visited nodes so that you
don’t need to do extra sorting at the end. And this will create a valid scheduling.
Proof of Correctness: Proof for cycle detection and topological sort will be the same as seen in
lecture. A directed graph is a DAG if and only if there is no back edge. And because DAG has no
back edge, if we order the vertices from highest post number to lowest post number, it’s guaranteed
to be topologically sorted by definition.
Runtime Analysis: Constructing the graph will take O(n + m) time since we have n nodes and m
directed edges. Running DFS will take again O(n + m). We can do linear scan to find back edge. If we
used stack to keep track of visited nodes, then topological sort will simply take linear time because we
can just read off the nodes from the back the stack. Therefore, the overall runtime will be O(n + m).

This content is protected and may not be shared, uploaded, or distributed. 5 of ??

CS 170, Fall 2024 Discussion 3 P. Raghavendra and S. Garg

5 Biconnected Components
Consider any undirected connected graph G = (V, E). We say that an edge (u, v) ∈ E is critical if
removing it disconnects the graph. In other words, the graph (V, E \ (u, v)) is no longer connected.
Similarly, we call a vertex v ∈ V critical if removing v (and all its incident edges) leaves the graph
disconnected.

(a) Can you always find a vertex v ∈ V that is not critical? What about an edge that is not
critical?

(b) Give a linear time algorithm to find all the critical edges of G.

(c) Modify your algorithm above to find all the critical vertices of G.

(d) A biconnected component of G is a connected subgraph that does not contain critical vertices;
in other words, if you remove a vertex from the component, then it will remain connected. If
we collapse all the biconnected components of G into a meta-node (similar to what we did with
SCCs), what does the resulting graph look like?

Solution:

(a) Consider running DFS from any vertex on the graph, and any leaf in the resulting DFS tree.
The leaf can be removed without disconnecting the graph, since the remaining vertices are
connected using the DFS tree edges. But we can’t always find such an edge. For example, every
edge in a tree is critical.
(b) Perform DFS on G while keeping track of pre[v] for each vertex. We also maintain a low value
for each vertex, where low[v] denotes the smallest pre[u] such that there is a back edge to u

This content is protected and may not be shared, uploaded, or distributed. 6 of ??

CS 170, Fall 2024 Discussion 3 P. Raghavendra and S. Garg

from the subtree of v. The only potential critical edges are the tree edges in the DFS. An edge
between v and its parent p is critical if and only if low[v] > pre[p] (i.e. there does not exist a
back edge from v to p or any of its ancestors).
(c) The root of the DFS tree is critical iff it has more than one child. A non-root vertex v is critical
iff for at least one of its children c, low[c] > pre[v].
(d) The resulting graph is a tree.
To prove this, show that the if there is a cycle in the resulting graph then all the meta nodes
in the cycle will belong to the same biconnected component. This gives a contradiction.

This content is protected and may not be shared, uploaded, or distributed. 7 of ??