CSE 347 Lecture 5

Reductions

1
Takeaway from Bipartite Matching
• We saw how to solve a problem (bi-partite
matching and others) by reducing it to another
problem (maximum flow).
• In general, we can design an algorithm to map
instances of a new problem to instances of a
known solvable problem (e.g., max-flow) to
solve this new problem!
• Mapping from one problem to another which
preserves solutions is called reduction.
2
 Reduction: Basic Idea
Convert solutions to the known problem to the
solutions to the new problem

Instance Instance Solution Solution

of new of known of known of new
problem problem problem problem

3
Reduction: Formal Definition
• Problems 𝐿, 𝐾

𝐾
Φ
𝐿 𝐾′

𝐿 reduces to 𝐾 (𝐿 ≤ 𝐾) if there is a mapping Φ from

any instance 𝑙 ∈ 𝐿 to some instance Φ(𝑙) ∈ 𝐾′ ⊂ 𝐾,
such that the solution for Φ(𝑙) yields a solution for 𝑙

4
Using reduction to design algorithms
In the example of reduction to solve Bipartite Matching:
𝐿:
𝐾:
Efficiency:
1. Reduction: Φ: 𝑙 → Φ(𝑙)
2. Solve problem Φ(𝑙)
3. Convert the solution for Φ(𝑙) to a solution to 𝑙

5
Efficient Reduction
A reduction Φ: 𝐿 → 𝐾 is efficient (𝐿 ≤𝒑 𝐾) if for any
𝑙 ∈ 𝐿:
1. Φ(𝑙) is computable from 𝑙 in polynomial (|𝑙|) time.
2. Solution to 𝑙 is computable from solution of Φ(𝑙) in
polynomial (|𝑙|) time.

We call 𝐿 is poly-time reducible to 𝐾, or 𝐿 poly-time

reduces to 𝐾.

6
Which problem is harder?
Theorem: If 𝐿 ≤" 𝐾 and there is a polynomial time
algorithm to solve 𝐾, then there is a polynomial time
algorithm to solve 𝐿.
Proof: Given an instance of 𝑙 ∈ 𝐿
1. Compute Φ(𝑙):
2. Solve Φ(𝑙):
3. Convert solution:
Total time:
Need to show:

7
Φ 𝑙 = 𝑝𝑜𝑙𝑦 𝑙 ?
Proof:

8
Hardness Problems
• Reductions show the relationship between
problem hardness!
Question: Could you solve a problem in polynomial
time?
Easy:
Hard:

9
Types of Problems
Decision Problem: Yes/No answer
• Examples:
1.
2.
3.
Optimization Problem: What is the value of an
optimal feasible solution of a problem?
• Minimization:
• Maximization:

10
Canonical Decision Problem
Does the instance 𝑙 ∈ 𝐿 (an optimization problem) have
a feasible solution with objective value 𝑘:

11
Hardness of Canonical Decision Problems
Lemma 1: 𝐷𝐿 ≤! 𝐿
Proof: Assume 𝐿 maximization problem
𝐷𝐿(𝑙): does 𝑙 have a solution ≥ 𝑘.
Let 𝑣 ∗ 𝑙 be the maximum objective on 𝑙 by solving 𝑙.

12
Hardness of Canonical Decision Problems
Lemma 2: If 𝑣 ∗ 𝑙 = 𝑂(𝑐 # ) for any constant 𝑐, then
𝐿 ≤! 𝐷𝐿.
Proof: First we could show 𝐿 ≤ 𝐷𝐿. Suppose maximization
problem, canonical decision problem is is there a solution ≥ 𝑘.
Naïve Linear Search: Ask 𝐷𝐿(𝑙, 𝑘), if returns false, ask
𝐷𝐿 𝑙, 𝑘 + 1 until returns true
Runtime:

This is exponential! How to reduce it?

13
 Our old friend Binary Search is back!

01 2 4 8 16

You gets a no at some value

# questions:
Binary search in area:
Runtime:

14
Reduction for Algorithm Design vs
Hardness
• Problems 𝐿, 𝐾

𝐾
Φ
𝐿 poly
𝐿′

If 𝐾 is “easy” (exists a poly-time solution), then

If 𝐿 is “hard” (no poly-time solution), then

Every problem that we worked on so far, 𝐾 is “easy”, so

we reduce from new problem 𝐿 to known problem 𝐾
(e.g., max-flow). 15
Reduction for Hardness: Independent
Set (ISET)
Input: Given an undirected graph 𝐺 = 𝑉, 𝐸
A subset of vertices 𝑆 ⊆ 𝑉 is called an independent set
if no two vertices of 𝑆 are connected by an edge.
Example:

16
Reduction for Hardness: Independent
Set (ISET)
Input: Given an undirected graph 𝐺 = 𝑉, 𝐸
A subset of vertices 𝑆 ⊆ 𝑉 is called an independent set if
no two vertices of 𝑆 are connected by an edge.
Problem: Does 𝐺 contain an independent set of size ≥ 𝑘?
𝐼𝑆𝐸𝑇 𝐺, 𝑘 returns true if 𝐺 contains an independent set
of size ≥ 𝑘, and false otherwise.
Algorithm? NO! We think that this is a hard problem.
A lot of people have tried and could not find a
poly-time solution

17
Pause to reflect

18
Example: Vertex Cover (VC)
Input: Given an undirected graph 𝐺 = 𝑉, 𝐸
A subset of vertices C ⊆ 𝑉 is called a vertex cover if 𝐶
contains at least one end point of every edge.
Formally, for all edges (𝑢, 𝑣) ∈ 𝐸, either 𝑢 ∈ 𝐶, or 𝑣 ∈ 𝐶.
Problem: 𝑉𝐶(𝐺, 𝑗) returns true if 𝐺 has a vertex cover of size
≤ 𝑗, and false otherwise
Example:

19
How hard is Vertex Cover?
Claim: 𝐼𝑆𝐸𝑇 ≤" 𝑉𝐶
Side Note:
DO NOT:
Proof: Show that 𝐺 = (𝑉, 𝐸) has an independent set of 𝑘
if and only if the same graph (not always!) has a vertex
cover of size 𝑉 − 𝑘.
Map:
𝐼𝑆𝐸𝑇 𝐺, 𝑘 → 𝑉𝐶( )

20
Proof of reduction: Direction 1
Claim 1: 𝐼𝑆𝐸𝑇 of size 𝑘 à 𝑉𝐶 of size 𝑉 − 𝑘
Proof: Assume 𝐺 has an ISET of size 𝑘: 𝑆, consider 𝐶 =
𝑉 − 𝑆, 𝐶 = 𝑉 − 𝑘
• Claim: 𝐶 is a vertex cover

21
Proof of reduction: Direction 2
Claim 2: 𝑉𝐶 of size 𝑉 − 𝑘 à 𝐼𝑆𝐸𝑇 of size 𝑘
Proof: Assume 𝐺 has an VC of size V − 𝑘: 𝐶, consider
S = 𝑉 − 𝐶, 𝑆 = 𝑘
• Claim: 𝑆 is an independent set

22
What does poly-time mean?
Algorithm runs in time polynomial to input size.
• If the input has 𝑛 items, algorithm runs in Θ(𝑛# ) for
any constant 𝑐 is poly-time.
– Examples: intervals to schedule, number of integers to
sort, # vertices + # edges in a graph
• Numerical Value (Integer 𝑛), what is the input size?
– Examples: weights, capacity, total time, flow constraints
– It is not straightforward!

23
Real time complexity of F-F?
In class: 𝑂(𝐹( 𝑉 + 𝐸 )
• 𝑉 + 𝐸 = this much space to represent the graph
• 𝐹: size of the maximum flow.

If every edge has capacity 𝑐, then 𝐹 = 𝑂(𝐶𝐸)

Running time: 𝑂(𝐶 𝐸 𝑉 + 𝐸 )

24
What is the actual input size?
Each edge ( 𝑬 edges):
20 10
• 2 vertices: |𝑉| distinct symbol,
20 log |𝑉| bits per symbol
s t • 1 capacity: log 𝐶
10 20 Size of graph:
• 𝑂(|𝐸|(|𝑉| + log 𝐶)
– 𝑃𝑜𝑙𝑦( 𝐸 , 𝑉 , log 𝐶)
Running time:
• 𝑃𝑜𝑙𝑦( 𝐸 , 𝑉 , 𝐶 )
– Exponential if 𝐶 is exponential
25
in 𝑉 + |𝐸|
Pseudo-polynomial
Naïve Ford-Fulkerson is bad!
Problem 𝐿’s inputs contain some numerical values, say
𝑊. We need only log 𝑊 bits to store 𝑊. If algorithms runs
in 𝑃𝑜𝑙𝑦(𝑊), then it is exponential, or pseudo-
polynomial.

In homework, you improved F-F to make it work in

𝑃𝑜𝑙𝑦( 𝑉 , 𝐸 , log 𝐶), to make it a real polynomial
algorithm.

26
Conclusion: Reductions
• Reduction
– Construction of mapping with runtime
– Bidirectional proof
• Efficient Reduction 𝐿 ≤" 𝐾
– Which problem is harder?
– If 𝐿 is hard, then 𝐾 is hard. à Used to show hardness
– If 𝐾 is easy, then 𝐿 is easy. à Used for design algorithms
• Canonical Decision Problem
– Reduction to and from the optimization problem
• Reduction for hardness
– Independent Set ≤! Vertex Cover 27