Projects

Projects for Analysis of Algorithms
Andreas Klappenecker
Project 1 (Cliques in Graphs). The Hamming distance dist(u, v) between two

binary vectors v = (v1 , . . . , vn ) and w = (w1 , . . . , wn ) is the number of indices k
such that vk 6= wk . A fundamental question in coding theory is to determine the
number
A(n, d) = max |{S ⊂ {0, 1}n | dist(u, v) ≥ d for all distinct u, v ∈ S}|,
the maximal number of binary vectors of length n that one can find such that any
two distinct vectors have a Hamming distance ≥ d. For example, A(5, 4) = 2.
The Hamming graph H(n, d) = (V, E) is the graph with 2n vertices V given by
binary strings of length n. We have (u, v) ∈ E if and only if dist(u, v) ≥ d.
The number A(n, d) coincides with the size of a maximal clique in H(n, d).
Find an implement “efficient” algorithms to compute the maximal clique in the
Hamming graph (but note that the problem to compute maximal cliques is NP
hard).
Background http://www.research.att.com/∼njas/doc/pace2.ps
Challenge http://www.research.att.com/∼njas/doc/graphs.html
Project 2 (Minimum Spanning Trees). Finding the minimum spanning tree of

an undirected connected graph (V, E) with weight function w : E → R is an old
problem in computer science. We have discussed the algorithms by Kruskal and by
Prim. The asymptotically fastest algorithm known to date is due to Chazelle. It
is an improvement of Borůvka’s algorithm, the first algorithm that was developed
for this problem. Extend the Boost graph library by Borůvka’s and Chazelle’s
algorithms. Compare the performance of the algorithms by Borůvka, Chazelle,
Kruska, and Prim.
Note that you should have some experience with C++ to attempt this project.
You can also decide to implement the algorithms without integration into the Boost
graph library, but the result will be less interesting.
Background Nesetril, Milková, and Nesetrilová on Borůvka’s algorithm
http://citeseer.ist.psu.edu/nesetril00otakar.html
Chazelle’s algorithm
http://www.cs.princeton.edu/∼chazelle/pubs/mst.pdf
Challenge Measure and document the performance of the implementations for
graphs that occur in interesting applications. Knuth’s Stanford graphbase is con-
tains numerous interesting examples.
1
Project 3 (The Traveling Salesman Problem, DIMACS Challenge). The travel-
ing salesman problem asks to determine the shortest tour for a salesman to visit
each city of a given list. Finding the best possible tour is NP-hard, but there ex-
ist surprisingly good heuristics. Take the DIMACS challenge, implement several
heuristics, and compare your results with the state of the art.
Background http://www.research.att.com/∼dsj/chtsp/
Challenge http://www.research.att.com/∼dsj/chtsp/download.html
Project 4 (Probabilistic Algorithms for k-SAT). k-SAT is a canonical NP-complete

problem. Recall that the SAT problem asks to determine whether or not a set of
clauses in conjunctive normal form is satisfiable. The problem is called k-SAT if
each clause has at most k literals.
Recently some progress has been made in designing randomized algorithms for
k-SAT. Two randomized algorithms are given below. Both of them are extremely
simple (a few lines of pseudo-code) and run in exponential time. The first one is
based on a backtrack search procedure and the second one is based on a random
walk. The second algorithm is superior asymptotically. You task is to determine
whether or not the same is true empirically. Implement both algorithms and com-
pare their performance in solving 3-SAT instances. Two sets of instances are pro-
vided. The first set consists of benchmark 3-SAT instances, and the second one
consists of randomly generated 3-SAT instances with unique solutions. Start with
small instances and report which is the largest instance that is solvable by your
program in a certain amount of time (you decide the time limit). Describe your
implementations and explain your findings in a way that is understandable to your
classmates.
Background
• Algorithm 1: R. Paturi, P. Pudlak, and F. Zane: Satisfiability coding lemma.
Proceedings of the 38th IEEE Symposium on the Foundations of Computer
Science, 1997, pages 566-574.
• Algorithm 2: Uwe Schöning: A Probabilistic Algorithm for k -SAT Based on
Limited Local Search and Restart. Algorithmica 32(4): 615-623 (2002)
3-SAT Challenge Instances
• Benchmark instances:
http://www.intellektik.informatik.tu-darmstadt.de/SATLIB/benchm.html
• Program that randomly generates unique solution 3-SAT instances :

http://www.is.titech.ac.jp/∼watanabe/gensat/a1/
2
Project 5 (Quadratic Sieve). The quadratic sieve is one of the first algorithms
to factor integers into primes in subexponential time. Implement the standard
quadratic sieve algorithm by Pomerance and take great care to realize the linear-
algebra part of the
√ algorithm efficiently. In the basic quadratic sieve, one chooses
integers x near n to search for values x2 − n that are B-smooth, √ i.e., that has
all factors in the range [1, B]. As the numbers x deviate from n, the B-smooth
numbers thin out rapidly. One way to get around this problem by choosing mul-
tiple polynomials, a method that has been suggested by Davis, Holdridge, and
Montgomery. Implement the multiple-polynomial quadratic sieve as well. Find the
right trade-offs between the size of the factor base, the number of polynomials, and
fine-tune the implementations.
Background R. Crandall, C. Pomerance: Prime Numbers – A Computational Per-
spective, Springer, 2001; Chapter 6.
C. Pomerance, The Quadratic Sieve Factoring Algorithm. In: Advances in Cryptol-
ogy: Proceedings of EUROCRYPT 84 (Ed. Th. Beth, N. Cot, and I. Ingemarsson).
pp. 169-182, Springer, 1985.
Challenge Factor integers such as
RSA-100 =
15226050279225333605356183781326374297180681149613\
80688657908494580122963258952897654000350692006139
(100 digits, checksum = 294805)
Project 6 (Matrix Multiplication). The multiplication of n × n matrices takes

O(n3 ) with the standard multiplication algorithm. We can improve the time com-
plexity of matrix multiplication to O(nω ), with ω = log 7/ log 2, using Strassen’s
well-known divide-and-conquer algorithm. Coppersmith and Winograd showed
that there exists a matrix multiplication algorithm that improves the exponent
to ω ≤ 2.38, but the algorithm has a prohibitively large constant and is appar-
ently not used in practice. Your task is to find and implement matrix multiplica-
tion algorithms that have a good performance for n × n matrices of, say, length
10 ≤ n ≤ 1000.
Background V. Strassen, Gaussian Elimination is not Optimal, Numer. Math.
13, p. 354-356, 1969
D. Coppersmith, S. Winograd, Matrix multiplication via arithmetic progressions,
J. Symbolic Comput. 9, p. 251-280, 1990
Challenge What is the largest n so that you can multiply two n × n matrices
within 1 second, 10 seconds, or 1 minute on a platform of your choice?
3
Project 7 (Shortest Vector). Suppose that a1 , . . . , an are n linearly independent
vectors in Qn . The lattice L generated by a1 , . . . , an is the set of all integer linear
combinations of these vectors, L = {λ1 a1 + · · · + λn an | λk ∈ Z for all 1 ≤ k ≤ n}.
The goal is to find the shortest nonzero vector in L with respect to the Euclidean
norm. In dimensions n = 1 and 2, the problem can be solved in polynomial time.
In higher dimensions, one can find approximate solutions using the LLL algorithm.
Background D. Micciancio and S. Goldwasser, Complexity of Lattice Problems:
A Cryptographic Perspective, Kluwer, 2002
V.V. Vazirani, Approximation Algorithms, Springer, 2001; see Chapter 27.
Challenge TBD
Project 8 (Network Reliability). Given a connected, undirected graph G = (V, E),

with failure probability pe specified for each edge e ∈ E, compute the probability
that the graph G becomes disconnected. Put differently, the graph becomes dis-
connected if all edges in some cut (C, C), with C ⊂ V , fail.
The project should find and implement algorithms which estimate the prob-
ability that the graph becomes disconnected. If each edge has the same failure
probability, then this is essentially a counting problem. An exact calculation of the
probability that the network partitions is a herculean task, since this problem is
]P-complete. Fortunately, there exist fully polynomial randomized approximation
schemes.
Background Some basic references are:
V.V. Vazirani, Approximation Algorithms, Springer, 2001; see Chapter 28.
http://theory.lcs.mit.edu/∼karger/Papers/reliability-sirev.ps
Challenge Obtain the data of real-world networks and calculate the probability
that the network partitions due to edge failures.

Projects

Uploaded by

Copyright:

Available Formats

Projects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Projects

Uploaded by

Copyright:

Available Formats

Projects for Analysis of Algorithms

Project 1 (Cliques in Graphs). The Hamming distance dist(u, v) between two

Project 2 (Minimum Spanning Trees). Finding the minimum spanning tree of

Project 4 (Probabilistic Algorithms for k-SAT). k-SAT is a canonical NP-complete

• Program that randomly generates unique solution 3-SAT instances :

Project 6 (Matrix Multiplication). The multiplication of n × n matrices takes

Project 8 (Network Reliability). Given a connected, undirected graph G = (V, E),

You might also like