Sunteți pe pagina 1din 6

Maad M.

Mijwel computer science, 2016

Travelling Salesman Problem


Mathematical Description
Maad M. Mijwel
College of Science, Baghdad University
Baghdad, Iraq
maadalnaimiy@yhaoo.com
December 2016

 found primarily in the forwarding sector (see also Logistics ), but


Abstract— The "traveling salesman problem" is about finding the also, for example, in connection with the definition of the order of
shortest round trip for a given number of cities. It looks simple, can the customers to be approached within the framework of direct
only be solved exactly for a small number of way stations, because sales, as carried out by some frozen food chains. To the solution
with every other city, the number of possible combinations This problem, which is fundamentally well structured as an integer
increases enormously: in ten cities, there are around 181,000 linear optimization problem, but extremely complex in terms of
possible routes that have to be calculated and compared. There are solvability, has proven itself to be small enough for the Branch &
already more than 1.8 million in eleven cities, and in 20 cities their Bound method ; for more realistic problem sizes, heuristics or
numbers are in the trillions range. newer approaches such as evolutionary algorithms or
local improvement methods (e.g. simulated annealing, tabu
Keyword— Traveling Salesman Problem, Symmetrical, search ) are more likely to be used (see also Operations ).
Metric, Graph, Cities.

I. INTRODUCTION II. MODELING AS A GRAPH

To be able to use mathematical methods for solution, a real situation

T he traveling salesman problem, Traveling Salesman Problem,


tour problem, traveling salesman problem, ordering problem is
a problem of operations research, which is to determine the
must first be mapped by a simple model. The problem of the
traveling salesman can be graphically modeled with the aid of a
graph, that is, by knots and edges. Here, the nodes (in the picture: A
optimal order of cities or machines, in which the total incurred
to D) represent the cities, while each edge ( )( ) between
kilometers, times or costs are a minimum , The problem owes its
two nodes and along with their length Depending on the
name to the problem of determining the optimal itinerary of
context, for example, it can be interpreted as the geographical length
a traveler who has to visit n places and must return to his place of
of a connection, travel time or the cost of a journey between two
departure at the end of the journey. It is a problem
cities. A tour (also called a Hamilton cycle) is a circle in this graph
of scheduling or sequencing, which is now in the planning of the
that contains every node exactly once. The goal is to find the
Production it matters. It can be solved by approximation
shortest possible tour. To simplify the study of the problem and to
methods. The Traveling Salesman Problem (TSP) is one of the
make sure that there is a tour, it is usually assumed that the graph is
classic optimization issues. It can be described as follows on the
complete, that is, there is always an edge between every
basis of a graph: A sales representative, starting in his hometown
two nodes. This can be achieved by inserting an artificial, very
(node 1), would like to visit n - 1 further customer locations (nodes
long edge wherever there is no edge. Due to its long length, such
2, ..., n) in an ascertainable sequence and then return to the starting
an edge will never occur in the shortest tour, unless there is no other
point. The total distance traveled should be minimal. So in the graph
tour. Depending on the properties of the edge weights, different
is a cycle ( round trip) with minimum length, which contains all
special cases of the problem are distinguished, the most important of
nodes exactly once. The TSP (in directed graphs) can be formulated
which are the symmetric and the metric TSP.
using binary variables xij that assume the value 1 if arrow (i, j) is in
the cycle and 0 otherwise, as shown in the next column. The first
two sets of constraints guarantee that each location will be reached
exactly once and left exactly once. The specified as short
cycle conditions and not formulated constraints ensure that
the round trip to be determined contains all nodes of the graph. In
particular, it is necessary to prevent short cycles with fewer than n
nodes. The short cycle with nodes 1, 2 and 3 are excluded, for
example. Overall, a large number of such conditions must be
formulated to avoid all conceivable short cycles. To a solution of
TSP diverse precise and exist heuristic method. The former are
mainly B & B procedures. With the latter, there are opening
procedures (e.g. nearest neighbor, cheapest insertion)
and improvement procedures (e.g. r-opt), which are Figure.1. Symmetrical TSP on 4 cities
often combined with meta-heuristics. That as a problem or
Traveling Salesman Problem describes a planning situation a) Symmetrical TSP
( planning ) in which to search for a minimum distance or transport In the general asymmetric TSP, the edges in the back and
cost, which includes a predetermined set of places to travel exactly forth direction can have different lengths, so this problem must be
once, and represents a closed cycle. This planning problem can be modeled using a directed graph. So it's not enough just to talk about
the connection between two nodes and their length, in addition, the
direction has to be specified. When symmetric TSP, however, is for

maadalnaimiy@yahoo.com
Maad M. Mijwel computer science, 2016

all node pairs )( the edge lengths are identical in both vector, shall be presented by way of example must meet in any
directions, it applies . case so that he describes a tour. They can, therefore, be used as part
of a relaxation in a cutting plane method to solve the problem.
As a consequence, each tour has the same length in both
directions. The symmetry halves the number of possible
tours. Asymmetric TSP is usually modeled using an undirected
graph (as in the picture). A traveling salesman problem between real
cities can be asymmetric or symmetrical, depending on whether, for
example, construction sites or one-way streets take longer in one
direction than the other or not.

b) Metric TSP
Asymmetric TSP is called metric if, in addition, its edge
lengths satisfy the triangle inequality. Clearly, this means that
detours are not worthwhile, because the direct connection of
after never is longer than the way of after over a third node : Figure.2. Degree condition: In every node, exactly one edge of the tour has
to go in and out.

Each node must be connected to the remaining nodes via exactly


Such edge lengths define a metric distance function on the node set, two tour edges, namely by an in and an outgoing edge :
that is, a distance measure that satisfies the conditions intuitively
expected from a distance. Several distance functions that occur
frequently in practice are metric, thus fulfilling the triangular
inequality : for all , In sum is every summand either 1 (included in the
 the Euclidean metric of the Euclidean TSP, tour) or 0 (not included). The sum, therefore, counts exactly the
number of edges of the tour that the node as end nodes. It must
 the Manhattan metric (also called the city block metric) of have the value 2 since one edge each has to lead in and out. In the
the rectilinear TSP, where the distance between picture on the right is a node shown with in and outgoing edges,
two nodes of a lattice graph (such as the Manhattan road the tour edges are marked in bold. At the edges are the values
network) is the sum of the x andy distances; that contribute to the above sums.
 or the maximum metric, where the distance between
two nodes of a grid-shaped graph is the maximum of the
distances in the x andy directions, respectively.
The last two metrics are used, for example, in the drilling of circuit
boards, where a drill, which has to process a given amount of holes
in the shortest possible time, can be moved independently in both
dimensions in order to move from one hole to the next. The
Manhattan metric corresponds to the case that the movement takes
place in both directions in succession, while in the maximum metric
both movements occur simultaneously and the total time is
determined by the respective longer distance in the x or y-direction. Figure.3. Short cycles: This variable assignment fulfills all degree
conditions, but does not define a tour.
Distances that are taken from distance tables in street maps are not
necessarily metric because there the km-length of an often time- the The above condition is fulfilled not only by tours but also by
shortest path is specified, which is not necessarily the shortest path variable assignments which describe several
with respect. The travel route. If it is permissible to visit places separate circles (so-called short cycles ), whereby each node is
several times in the practical planning problem, one can reduce the contained in exactly one circle.
problem to a metric TSP in the so-called distance graph. There, the To exclude this, short cycle inequalities (also called sub-tour
edge length corresponds to two nodes the length of the elimination conditions) have to be fulfilled. They say that any node
shortest Path with respect to the original edge lengths , The set that is neither empty nor contains all nodes must be connected to
new edge lengths defined in this way always fulfill the triangle the remaining nodes by at least two edges of the tour :
inequality.

III. MODELING AS AN INTEGER LINEAR PROGRAM for all node sets with. The total counts all edges of the Tour between
One approach to solving the problem is to formulate it as an integer a node and another node . To avoid
linear program that describes decisions by variables and conditions redundant inequalities, one can also rely on node sets with at least
by linear inequalities. There are several possible variants for two and at most nodes . In the picture on the right,
this. By way of example, a modeling for the symmetrical TSP node the edges { }with bold, while the remaining edges the
set is intended here to be introduced. For value . Adding condition (2) for the node-set consisting of
every edge becomes a binary variable introduced,
the three left nodes would ensure that must be connected to the
which indicates for a given tour, whether the edge included in three right nodes through at least two tour edges, thereby excluding
this tour or not, Each tour can be specified in this way by the two short cycles shown
specifying the associated variable values, but not every 0-1 The number of inequalities of type (2) grows exponentially with the
assignment of the variable values defines a tour. The conditions for number of cities since almost every one of the Subsets of nodes
a variable assignment to define a tour can be expressed by defines an inequality. However, this problem can be circumvented
linear inequalities, although not all of them are known. In the
following, therefore, two simple types of inequalities, which are a

maadalnaimiy@yahoo.com
Maad M. Mijwel computer science, 2016

by using cut-plane techniques that add these inequalities only when


they are actually needed.
The goal is to find a shortest of all tours:

Since the variables, it can be shown that most of the conditions (1)
and (2) define facets of the TSP polytope, that they belong to the
best linear inequalities that can be used to describe a tour. However,
they are not enough, there are vectors .

IV. ALGORITHMIC COMPLEXITY


When looking for a solution to the traveling salesman's problem,
one of the most fundamental questions is the number of possible Figure.4. Section plane intersects an invalid solution (blue point ) from the
round trips. In each node of the tour, the salesman can choose from red polytope of the permissible integer points. The next iteration will yield
all the cities he has not yet visited. Since the starting point is the green point, which is closer to the red polytope than the cut blue point.
arbitrary, total results .
The traveling salesman's problem is NP-complete, even if a Geometrically interpreted, these methods consider the problem
certain metric is required. This prevents the existence of a as a convex polytope, thus as a multidimensional polygon in then
polynomial solution algorithm under the as yet unproven but -dimensional space, where the number of edges of the graph
generally assumed assumption that the two complexity classes P and is. This polytope is part of the unit cube from
NP are not identical. Illustratively, this means that the duration of which inequalities cut off parts (like a piece of cheese). As
any deterministic algorithm for optimally solving this problem described above, each vertex of this unit cube (that is, each vector
increases exponentially with the number of cities so that such consisting of 0/1 entries only) represents a tour, provided that
an algorithm From a theoretical point of view "not essential" can be it satisfies certain linear inequalities.
better than trying out all kinds of tours, which is not feasible even The described inequalities thus cut off exactly the corners of the
with small city numbers. For example, there are more than 43 unit cube that do not represent a tour. For example, inequalities of
billion possible roundtrips for the symmetrical TSP with 15 cities, type (1) intersect inter alia the zero vector (0, ..., 0) that would
with 18 cities already over 17 trillion. correspond to an (impossible) tour without edges. What remains is a
polytope in which every integer corner describes a tour. The
In addition, the search for good solutions is hampered by the
goal is, therefore, a 0/1 corner of with minimum objective
traveling salesman problem. (assuming P NP) can be
function value (3). the adjoining image shows the
approximated as easily as possible in polynomial time. So not find
any arbitrarily small polynomial algorithm ALG, which polytope an attempt is made to determine an optimal
calculated a tour for each TSP problem whose length is at most a corner of the inner polytope. The picture shows a green
factor deviates from the optimal value : cutting plane that shows the impermissible blue dot of the
inner polytope . The sole application of these methods is usually
indicates the length of an optimal tour, that of not enough to quickly find good round trips. Their main advantage
the algorithm calculated by the algorithm. However, two heuristics is that they provide information on how long the shortest tour must
have already been presented earlier for metric TSP and certain a be at least. With such a lower bound for the optimal solution value,
quality guarantee of namely the minimum spanning tree it can be estimated how good a found tour is in comparison to an
heuristic and the Christofias's heuristic. so far no algorithm optimal round trip, without knowing it. If for example, you have
with a better quality guarantee than ½ is known, just as little as one found a lower bound of 100 and a tour of length 102, the
approximate algorithm for general, non-metric TSP. so-called optimality gap is the relative distance between the lower
bound and shortest known tour length, (102-100) / 100 = 2%. Since
an optimal tour can only be between 100 and 102, the found solution
value 102 can also be at most 2% away from the optimum value. If
V. SOLUTION PROCESS METHODS
the length of a found tour is the same as the lower bound, it proves
The known solution methods are divided into two groups, which that the solution found is optimal. To find good tours, these exact
can be combined with each other. Exact solution methods - procedures can be combined with heuristics, some of which are
assuming any length of time required - basically a verifiable optimal described in the following section.
solution. Heuristic methods often find good solutions in a short
time, but in the general case, they can be as bad as they can be. For
the metric TSP, however, there are polynomial heuristics whose VI. HEURISTICS
solutions are generally no more than a factor of 1.5 or 2 longer than
the shortest round trip. A precise procedure, but due to the many In order to arrive quickly at useful solutions, heuristics, i.e.
possible tours is practically not feasible, is the enumeration of all approximation methods, are usually necessary, but as a rule, they do
possible tours. not provide an estimate of the quality of the solutions
On the other hand, methods of integer linear optimization, in found. Depending on whether a heuristic constructs a new tour or
particular, branch-and-cut , can demonstrably optimally solve attempts to improve an existing itinerary, it is referred to as
instances of practically relevant magnitudes, or at least estimate the an opening (or construction) or improvement process.
quality of a found tour in comparison to an optimal solution. a) Opening procedure
The intuitive approach of a traveling salesman probably
corresponds most closely with the nearest neighbor heuristic (next
neighbor). Starting from a city, this selects the nearest as the
following location. This will be continued successively until all

maadalnaimiy@yahoo.com
Maad M. Mijwel computer science, 2016

cities have been traveled and the traveling salesman has returned to calculated, which is at most one and a half times as long as an
the starting point optimal one. Here, instead of doubling the edges in
the MST heuristics, a least perfect mating is calculated on
the odd-numbered nodes in the minimally spanning tree to
produce a Eulerian graph. However, this algorithm is more
complex.

b) Enhancement procedures
Improved optimization methods, also Post-Optimization
(post-optimization) try to shorten an existing tour through small
modifications. If none of the considered changes leads to an
improvement, then a local optimum is found (but not necessarily a
global one).
Figure.5. Nearest Neighbor Algorithm The heuristics pursue this approach by
Starting from a city, this selects the nearest as the following
location. This will be continued successively until all cities have systematical groups of replacing so that again a tour arises. The
been traveled and the traveling salesman has returned to the starting problem with such methods, however, quickly becomes the fact that
point. In every city, the shortest outgoing route has to be in a complete implementation of the effort compared to the list of all
searched. At most, there can only be as many outgoing edges per possible tours would not be lowered. In practical implementations
city as nodes are present in the graph. This results in an algorithmic used The quality of a heuristic in practice strongly
complexity of O (n²), so the number of computation steps depends
on quadratic ally on the number of cities. However, the fact that this depends on the selection of the edges and the parameter to
heuristic generally does not provide the best solution is due to the be exchanged
fact that the distance between the starting city and the last visited
city is not taken into account until the very end. The nearest
neighbor heuristic can yield arbitrarily poor results, i.e. there is no c) Metaheuristic procedure
constant, instance-independent approximation factor for the
solution value compared to the optimum value. Metaheuristics combine local and global search techniques in an
A whole class of further opening procedures forms the abstract strategy for the heuristic optimization of a problem. Many
so-called insertion heuristics. The simplest variants of this are the of these methods calculate a heuristic starting solution (for example,
nearest insertion heuristic (next insertion) and the farht insertion with the nearest neighbor heuristic) and improve it by a local search
heuristic (farthest insertion). Given are (few) neighboring cities for method, such as. B. K-Opt heuristics until no better tour is
which an optimal round trip can be quickly determined by exact found. Tab-Lists prevent already found tours from being viewed
procedures. Now it is checked step by step, which is not yet visited again. To prevent the sticking in local minima, you can, for example
city closest (or farthest) to one of the connecting lines of the
previous round trip. If this city is found, it will be built into the tour
 start several experiments (mountaineering algorithm),
of the cities nearest to it. The process continues until the round trip  in between accept larger and later only minor
includes all cities. The solutions of this heuristic can be arbitrarily deteriorations (simulated annealing (also known as the
bad compared to an optimal solution. deluge algorithm)),
Another class of heuristics divides the node set into
individual partitions (e.g., according to geographic criteria), each of  combine several already found tours to a new tour or
which is partially optimized. Subsequently, the partial solutions are change it randomly (evolutionary and
combined to a total solution. As a rule, this is only locally optimal genetic algorithms ).
and can be arbitrarily poor compared to the global optimum.
A whole class of other nature-inspired methods uses swarming
The minimum spanning tree heuristic ( MST ) first computes a intelligence. In Ant Colony Optimization (Ant Colony
minimal spanning tree, i.e. a graph in which all points are connected Optimization), this model is used to model the natural behavior of
to each other and has the minimum length. Based on this, a tour is ants on the way, while Particle Swarm Optimization (" Particle
constructed by first doubling all tree edges and then searching for Swarm Optimization”) uses the behavior of bird or fish swarms as
a Euler tour in the resulting Eulerian graph. This is an example. Artificial neural networks and similar statistical
last abbreviated by direct edges if nodes are visited twice. In the methods can also be used to efficiently find useful solutions. Models
case of a metric TSP, one can show that the tour so constructed is at such as the Hopfield network or Self-Organizing Maps can map
most twice as long as the shortest tour. many aspects of the TSP. The big disadvantage of artificial neural
networks, however, is that often only local minima is found.
In general, the success and duration of metaheuristic procedures
depend significantly on the definition and implementation of the
individual steps. In principle, all these methods can calculate good
solutions, but also arbitrarily poor compared to an optimal solution,
unless used to calculate the starting solution, a heuristic with quality
guarantee and their value is then never exceeded later. But you can
usefully z. B. be used in the context of a branch-and-cut algorithm.

d) Dual heuristics
The traveling salesman's problem is one of the few combinatorial
optimization problems in which it is easy to specify useful lower
Figure.6. Low spanning tree bounds for the minimum length of a tour (in general, the minimum
cost of a solution). For example, every tour, even an optimal,
An even better approximation quality for metric TSP is achieved
by the Christofias's heuristic. With her, a round trip can be

maadalnaimiy@yahoo.com
Maad M. Mijwel computer science, 2016

accurate Contains circles. The tour is at least as long as [8] Baltz, A., and A. Srivastav, Approximation algorithms for the
Euclidean bipartite TSP, Operations Research Letters, 33 (2005)
this tree and by definition also at least as long as a minimally 403-410.
spanning tree (i.e. a spanning tree with a minimum sum of edge [9] Barvinok, D. Johnson, G. Woeginger, R. Woodroofe, The
lengths) that can be easily determined. Since this applies to each maximum traveling salesman problem under polyhedral norms,
tour, the length of a minimally spanning tree provides a lower Proceedings of the 6th IPCO Conference on Integer Programming
bound for the length of an optimal tour. and Combinatorial Optimization, 1998, p. 195-201.
[10] Barvinok, A. E. Gimadi, A. Serdyukov, The maximum
traveling salesman problem, in The Traveling Salesman Problem
and its Variations, G. Gutin and A. Punnan, (eds.), Kluwer,
VII. CONCLUSION Dordrecht, 2002, p. 585-607.
The largest instance of a (symmetric) round trip problem, which [11] Bianco, L. and A. Mingozzi, S. Ricciardelli, Dynamic
has so far been proven to be optimally solved, is a planning problem programming strategies and reduction techniques for the traveling
for the layout of integrated circuits with 33,810 nodes. This record salesman problem with time windows and precedence constraints,
was created in 2005 by William Cook and others using Operations Research, 45 (1997) 365-377.
a combination of different heuristics and the branch-and-cut [12] Charikar, M. and B. Raghavachari, the finite capacity
based Concorde program using previous partial results of various dial-A-ride problem, In Proceedings of the 39th Annual IEEE
university working groups as a basis. The hitherto largest optimally Symposium on the Foundations of Computer Science, 1998.
resolved instance consisted of 24,978 Swedish cities, dissolved in [13] Chiang, Y-J, New approximation results for the maximum
2004. With the help of special decomposition techniques and the use scatter TSP, Algorithmica, 41 (2005) 309-341.
of several parallel computers, William Cook u. a. Found solutions [14] Christofides, N., Worst-case analysis of a new heuristic for the
for a TSP of over 526 million stars, which are proven to be at most traveling salesman problem, Report 388, Graduate School of
0.798% off the optimum. Industrial Administration, Carnegie Mellon U, 1976.
[15] Clarke, G., and J. Wright, Scheduling vehicles from a central
However, the fact that a TSP of a certain size could be solved
depot to a number of delivery points, Operations Research 12 (1964)
optimally does not imply that every instance of this size can be
568-581.
solved optimally because, as with most combinatorial optimization
[16] Dantzig, G. and J. Ramser, The truck dispatching problem,
problems, the difficulty of the solution is greatly different from the
Management Science 6 (1959) 80-91.
input parameters (in this case the edge weights). For TSPs that arise
[17] Deineko, V. and G. Woeginger, The convex-hull-and-k-line
from practical applications, additional constraints, such as time
traveling salesman problem: a solvable case, Information
windows, often need to be considered. Therefore, usually the largest
Processing Letters, 59 (1996) 295-301.
optimally solvable problem instances of such variants are
[18] Deineko, V. and M. Hoffmann, Y. Okamoto, G. Woeginger,
significantly smaller than the classic problem of the traveling
The traveling salesman problem with few inner points, Proceedings
salesman, so that in practice often resorting to heuristic approaches
of the 10th International Computing and Combinatory Conferences,
to the solution. combinations heuristic methods using LP-based
LNCS Volume 3106, Springer-Verlag, 2004, 268-277.
methods such as branch-and-cut are used primarily in the research
[19] Deineko, V. and R. van Dal, G. Rote, The convex-hull-and-line
environment in order to be able to assess the quality of solutions and
traveling salesman problem: a solvable case, Info. Proc. Lett., 59
solution methods with the aid of lower barriers for the tour length.
(1996) 295-301.
[20] Fekete, S. Simplicity and harness of the maximum traveling
ACKNOWLEDGMENT salesman problem under geometric distances, Proceedings of the
10th ACM-SIAM Symposium on Discrete Algorithms, 199, p.
I would like to thank my wife Aşkım for standing beside me 337-345.
throughout writing this article. [21] Flood, M., The traveling-salesman problem, Oper. Res. 4
(1956) 61-75.
REFERENCES [22] Gamarnik, D., and M. Lewenstein, M. Sviridenko, an improved
upper bound for the TSP in cubic 3-edge connected graphs,
[1] Applegate, D., and R. Bixby, V. Chvatal, W. Cook, TSP cuts Operations Research Letters, 33 (2005) 467-474.
which do not conform to the template paradigm, In Computational [23] Garfinkel, R, Minimizing wallpaper waste, Part I: a class of
Combinatorial Optimization, Optimal or Provably Near-Optimal traveling salesman problems, Oper. Res. 25 (1977) 741-751.
Solutions, LNCS, Volume 2241, Springer-Verlag, 2001, p. 261-304. [24] Garfinkel, R. and K. Gilbert, the bottleneck traveling salesman
[2] Arkin, E., and M. Bender, J. Mitchell, S. Skiena, the lazy problem: algorithms and probabilistic analysis, J. Assoc. Comput.
bureaucrat scheduling problem, Proceedings of the 6th International Mach., 25 (1978) 435-438.
Workshop on Algorithms and Data Structures, 1999, p. 122.-133. [25] Garey, M. and D. Johnson, Computers and Intractability, W. H.
[3] Arkin, E., and Y-J. Chiang, J. Mitchell, S. Skiena, T-C. Yang, Freeman, New York, 1979.
On the maximum, scatter traveling salesperson problem, SIAM J. [26] Grigni, M. and E. Koutsoupias, C. Papadimitriou, An
Computing, 29 (1999) 515-544. approximation scheme for planar graph TSP, In, Proc. IEEE
[4] Arora, S., Polynomial time approximation schemes for Symposium on the Foundations of Computer Science, 1995,
Euclidean traveling salesman and other geometric problems, Journal 640-645.
of ACM, 45 (1998) 753-782. [27] Gutin, G. and A. Punnen, (eds.), The Traveling Salesman
[5] S. Arora, and M. Grigni, D. Karger, P. Klein, A. Woolsey, a Problem and Its Variations, Kluwer, Nowell, MA., 2002.
polynomial-time approximation scheme for weighted planar graph [28] D. Johnson and L. McGeoch, The Traveling Salesman
TSP, Proc. 9th Annual ACM-SIAM Symposium on Discrete Problem: A Case Study in Local Optimization, in Local Search in
Algorithms, 1998, p. 33-41. Combinatorial Optimization, E. H. L. Aarts and J.K. Lenstra (ed),
[6] Arora, S. and C. Lund, R. Motwani, M. Sudan, M. Szegedy, Wiley, 1997, p. 215-310.
Proof verification and hardness of approximation problems, Journal [29] Karuno, Y. and H. Nagamochi, T. Ibaraki, A
of the ACM, 45 (1998) 501-555. 1.5-approximation for single-vehicle scheduling problem on a line
[7] Ausiello, G. and E. Feuerstein, S. Leonardi, L. Stougie, M. with release and handling times, In Japan-U.S.A. Symposium on
Talamo, Algorithms for the on-line traveling salesman, Flexible Automation, July 1998, p. 1363-1366.
Algorithmica, 29 (2001) 560-581.

maadalnaimiy@yahoo.com
Maad M. Mijwel computer science, 2016

[30] Kruskal, J., On the Shortest Spanning Subtree of a Graph and [41] Reinelt, G. "TSPLIB--A Traveling Salesman Problem
the Traveling Salesman Problem, Proc. Amer. Math. Soc. 7 (1956) Library." ORSA J. Comput. 3, 376-384, 1991.
48-50. [42] Rosenkrantz, D. J.; Stearns, R. E.; and Lewis, P. M. "An
[31] Lawler, E., A solvable case of the traveling salesman problem, Analysis of Several Heuristics for the Traveling Salesman
Math. Programming 1 (1971) 267-267. Problem." SIAM J. Comput. 6, 563-581, 1977.
[32] Lawler, E., and J. Lenstra, A. Rinnooy Kan, D. Shmoys, (eds.) [43] Shmoys, D. and D. Williamson, Analyzing the Held-Karp TSP
The Traveling Salesman Problem: A Guided Tour of Combinatorial bound: A monotonicity property with the application, IPL 35 (1990)
Optimization, Wiley, New York, 1985. 281-285.
[33] Lewenstein, M. and M. Sviridenko, A 5/8-approximation [44] Toth, P. and D. Vigo, (eds.), The Vehicle Routing Problem,
algorithm for the maximum asymmetric TSP, SIAM J. Discrete SIAM, Philadelphia, 2002.
Math., 17 (2003) 237-248. [45] Trevisan, L., When Hamming meets Euclid: the
[34] Lin, S. "Computer Solutions of the Traveling Salesman approximability of geometric TSP and Steiner tree, SIAM J.
Problem." Bell System Tech. J. 44 (1965) 2245-2269. Computing, 30 (2000) 475-485.
[35] Martello, S. and P. Toth, Knapsack Problems: Algorithms and [46] Tsitsiklis, J., Special cases of the traveling salesman and
Computer Implementations, Wiley, Chichester, 1990. repairman problems with time windows, Networks, 22 (1992)
[36] Miller, C. and A. Tucker, R. Zemlin, Integer programming 263-282.
formulations and traveling salesman problems, J. of the ACM, 7 [47] Woeginger, G., Exact algorithms for NP-hard problems: a
(1960) 326-329. survey, in Combinatorial Optimization - Eureka! You shrink!
[37] Papadimitriou, C., The Euclidean TSP is NP-complete, Lecture Notes in Computer Science, Vol. 2570, 2003, p. 185-207.
Theoretical Comput. Sci., 4 (1977) 237-244. [48] Maad M. Mijwel. Genetic Algorithm Optimization by
[38] Papadimitriou, C., The complexity of the Lin-Kernighan Natural Selection. Augusts,2016.
heuristic for the traveling salesman problem, SIAM J. on
Computing, 21 (1992) 450-465.
[39] Papadimitriou, C., and S. Vempala, On the approximability of
the traveling salesman problem (extended abstract). Proceedings of
STOC 2000, p. 126-133.
[40] Platzman, L. K. and Bartholdi, J. J. "Spacefilling Curves and
the Planar Traveling Salesman Problem." J. Assoc. Comput. Mach.
46, 719-737, 1989.

maadalnaimiy@yahoo.com

S-ar putea să vă placă și