You are on page 1of 4

This paper will summarise selected articles on the topic of the Hamiltonian path problem

and will offer a new proposal such as Björklund et al's. algorithm with O*(1.657n) time to solve
the specified problem of how to get to destinations most effectively.
The problem involves planning a multi-stop tourist path, or designing a route for a
rubbish collection truck is attributed to the Hamiltonian path problem. Such problems are also
related to the Travelling salesman problem (TSP), and both are NP hard problems. Fortnow’s [1]
article gives an overview of the P versus NP problem. Though he did not define the P and NP
problems, Fortnow defined which class of problem is related either to P (Polynomial Time) or to
NP (Nondeterministic Polynomial Time) problems. The article delivers historically notable
changes in this scientific field, which encourage the reader to clarify the meanings of P=NP or
P!=NP, and underlines the significance of each case. Although the article is presented in quite an
informal style, it gives sufficient information as it can help the reader to understand the nature
and complexity of the problem. It becomes clear that the Polynomial time algorithm cannot be
implemented for the given problem, otherwise it would imply that P=NP.
In addition, Fortnow [1] analysed techniques that attempted to settle the P versus NP
problem. He characterised the uniqueness and weaknesses of each technique. For example,
Diagonalization cannot settle the problem, because it requires simulation; the outcome of the
fixed ‘NP machine’ in the arbitrary ‘P machine’ is obscure [1]. Similarly, the approach using
tautologies of Disjunctive Normal forms does not have a short resolution proof. He concluded
this part of his paper by showing that the proof systems such as cutting planes, algebraic proof
systems and Frege axioms have weaknesses, according to complexity theorists.
The article by Fortnow explains the significance of P and NP problem and answers
questions why P is not equal to NP and vice versa. He also provides a wide range of techniques,
fast algorithms and methods for solving NP complete problems. Demonstrating weaknesses of
each method, he makes it clear that there is no polynomial algorithm that solves the NP hard
problems. These fully developed algorithms run in exponential time with approximations in
results or they can solve moderately sized problems. However, owing to the rapid development
of computers, the NP complete problems can be solved practically. The enormous number of
emerging algorithms can solve some of the NP complete problems with a small approximation;
therefore some of them have voluminous results. For instance, the Travelling salesman problem,
which asks us to find the tour with the shortest distance between cities visited once, can be
tackled by the combined algorithms with some restrictions. The problem can be solved by using
the cutting–plane method with some additions, and the result was over ten thousand cities. This
result was increased several times by other algorithms such as the triangle inequality or Approx-
TSP-Tour algorithm as well. Moreover, there is an efficient algorithm developed by Arora which
has the best possible route with a small approximation [1]. The comparison of all these
algorithms is complex because they use different methods. They require some properties such as
indirect or direct graph or cubic graph therefore their settings differ from each other. It is hard to
define the best one; the only attribute which can distinguish one algorithm from another are
running time and input size.
The best solution for the Hamiltonian path was established as O*(2n) running time in the
1960’s by different academics (Bellman, Held and Karp) [2]. Belman’s “dynamic Programming
treatment of the Travelling salesman problem” provides specific information to understand the
origin of the TSP problem and the method to solve it. In his notes he attempts to formulate the
given problem in dynamic programming terms. This method can solve the problem for up to 17

1
cities. However, applying additional simple manipulations to this method can increase the
number of cities and perform more quickly with approximation [2]. The current results for the
TSP problem are many times greater than the latest one.
The importance of Bellman’s article is to understand the core of the problem in terms of
programming. Although the article can be seen as out-of-date, the complexity and description of
the problem is unchanged. Bellman considered an indirect path with multiple iterations. Starting
from some fixed 0 city, this path goes to the city n and before returning to the starting 0 city
there are n {m1, m2, …, mk}cities which need to be visited, Bellman’s formula resulted in
optimality of route length, because the route from i through m1, m2, …, mk cities needs to be of
the minimum length. The iteration repeats through n to m1, m2, …, mk cities every time, choosing
the shortest one among the tour’s lengths. The obtained result relates to ‘rapid access’; however,
with a sufficient amount of time the size of the problem can increase by auxiliary storage
methods [2]. Storing function f (n; m1, m2, …, mk,) in the table of values of f(n; m1, m2, …, mk) will
give a table with all chosen tours K. The result for 11 cities has an approximation of less than
300, whereas for 21 cities there was a higher approximation of 200,000. Every iteration is a
comparison of a path that has n! size, however, Bellman reduced the number of comparisons to
n22n-1. Despite the limitation of computer resources (memory and storage space) at that time, the
result was impressive [2], as it explains in depth the principles of solving the Travelling
Salesman problem.
Since that time the computers have developed enormously, leading to the emergence of a
huge number of fast algorithms. Even those algorithms with some small approximation can solve
and optimize many real world problems. For example, in “Introduction to algorithms” [3]
alternative algorithms are given to solve the complete TSP problem. According to Cormen et al.
[3] the problem can be tackled practically through the implementation of triangle inequality. In
this case the algorithm tries to find the tour with the minimum cost. The path from u to w has
cost C(u,w). It is clear that the most direct path will be cheaper than with intermediate stops.
Therefore, cutting intermediate vertexes will not raise the total cost of the tour. Although the cost
function satisfies triangle inequalities, the method cannot determine an exact solution. To find an
exact solution, Cormen et al. [3] offers the use of the Approx-TSP-Tour algorithm. The
disadvantage of this method is the necessity to delete any visits to vertices that are visited more
than once. Therefore Cormen et al. conclude that there are other practical algorithms which
perform better than the Approx-TSP-Tour. This book covers not only NP complete or NP hard
problems but also some simple algorithms that can solve many moderate problems. The book [3]
is a guide as much for beginner programmers as for professionals. It includes descriptions of
problems, methods of solving them, and some exercises to consolidate gained knowledge.
Another alternative algorithm for solving the given problem is Björklund et al’s [4]
method. Their method can detect and report whether an undirected graph on n vertices has or
does not have a Hamiltonian path in O*(1.657n) time. This time is the fastest since Bellman
established the fastest time in O*(2n) in the 1960s. Since then the time has improved several
times. Broersma et al. [4] has O*(1.682n) time algorithms that detect Hamiltonicity in claw free
graphs. Iwama and Nakashima in cubic graphs have proved the TCP problem in O*(1.251n)
time; Gebauer’s algorithm counting the Hamiltonian cycles has O*(1.715n) [4]. Another
approach is to determine the time to find the noncrossing k path in an n vertex graph which has a
time 2knO(1) presented by Koutis and Williams [4].

2
In the article “Determinant Sums for Undirected Hamiltonicity”, Björklund et al. [4]
introduced the Monte Carlo algorithm. When applying this algorithm to solve the given problem
it is important to understand every step. In the article an explanation of the algorithm is given in
detail. The Monte Carlo algorithm is a combination and extension of some previous methods.
For example, the algorithm for Hamiltonicity used a similar approach to the inclusion-exclusion
algorithm. The inclusion-exclusion algorithm has some required characteristics, such as input
with polynomial space; the graph needs to be directed, deterministic and count for solutions,
whereas the Monte Carlo algorithm requires only polynomial space in the input. Unlike the
inclusion-exclusion algorithm, the Monte Carlo algorithm requires that the graph needs to be
undirected and no relation to the number of results. Distinction from the inclusion-exclusion, the
Monte Carlo algorithm does not count weighted cycle covers. Similarly to the algorithm of
Koutis and Williams, estimate polynomials with more variable quantities over fields filter for the
Hamiltonian cycles. The origin of the problem is using the determinant to count weighted cycle
cover [4].
In the Monte Carlo algorithm Björklund et al. reduced Hamiltonicity to the Labeled
Cycle Cover Sum; a reduction is a process which transforms the NP hard problem into a smaller
problem without changing the optimality of the NP hard problem. Defining their Labeled Cycle
Cover Sum, Björklund et al. stated that “The name stems from the fact that every arc in the cycle
cover is labeled by a nonempty subset of a set of labels” [4]. If the Labeled Cycle Cover Sum
results 0 there is no Hamiltonian cycle, otherwise there is at least one Hamiltonian cycle.
Björklund’s method using a determinant of the n x n matrix can detect the Hamiltonian cycle
comparably quickly. The strategy is to divide vertices of graph G into two similar halves, where
G is an undirected graph. The first half will compose graph D with two directions and the second
half is used as labels of Labeled Cycle Cover Sum on D. The presenting algorithm has a required
property, such as Polynomial space, and strongly depends on the indirectness of the graph. The
originality of the algorithm is in its use of determinants to count excellent matches.
Planning a multi-stop tourist path, or designing a route for a rubbish collection truck are
problems related to the Hamiltonian path problem or the Travelling Salesman Problem. The
Hamiltonian problem is a path in an undirected or directed graph that passes through vertices of
the graph only once. The Travelling Salesman Problem has an extension that also requires the
minimum length of the path. Both problems are NP hard problems. Therefore, to solve the NP
hard problem it needs to be reduced to a solvable problem, which means being converted by a
polynomial time algorithm.
Thanks to the Clay Mathematics Institute’s million-dollar prize, there is a sufficient
number of algorithms to solve the problem. There is a variation of the algorithms that can be
implemented to solve the given problem. Moreover, the algorithms resulting with approximation
are very close to the solution and the input size is significant. To solve the given problem, only
one of them can be implemented. As is known from Björklund et al’s research paper [4], the
previous best solution for TSP on the worst case, was O*(2n), from over 50 years ago. However,
since then computer power and storage availability has both exponentially grown leading to the
emergence of a number of algorithms with more approximation. The latest and best known
solution for the Hamiltonian path and TSP problem is Björklund et al’s [4] Monte Carlo
algorithm running in O*(1.657n) time and it can be applied with great success. However, in the
present era an ambitious researcher may well be able to prove his or her theorem with a better
running time.

3
References

[1] Fortnow, L. (2009). The status of the P versus NP problem. Communications of the ACM,
52(9), p.78.
[2] Bellman, R. (1962). Dynamic Programming Treatment of the Travelling Salesman Problem.
Journal of the ACM, 9(1), pp.61-63.
[3] Cormen, T., Leiserson, C., Rivest, R. and Stein, C. (2014). Introduction to algorithms.
Cambridge, Massachusetts: MIT Press.
[4] Björklund, A. (2014). Determinant Sums for Undirected Hamiltonicity. SIAM Journal on
Computing, 43(1), pp.280-299.

You might also like