Everything began in the haze of the 70's of the last millennium. It is nebulous who stumbled first upon this enigma. Some say that Ron Graham has discussed the problem in public lectures, some others say that Joseph O'Rourke has posed it as an open question in the American Mathematical Monthly, while some others suspect that the problem had been already hiding in older different formulations. However, it is a historical fact that one shinny/cloudy day, three computer scientists finished polishing a manuscript that became a classical paper known as "Some NP-complete geometric problems". In this paper, Michael Garey, Ron Graham and David Johnson showed the NP-hardness of two important problems in geometric metrics: Steiner Tree and Traveling Salesman. For the Euclidean plane, they showed only NP-hardness as they did not manage to show that these problems are contained in NP. Moreover, they accentuated that we cannot even rule out that the decision version of Euclidean Minimum Spanning Tree is outside of NP. What a seeming paradox given that we can compute such a minimum tree in polynomial time! So, whom did they blame? The short answer: The Euclidean metric. Garey and his coauthors explain that all these problems have a common hidden issue: They rely on comparing Euclidean lengths, that is, they rely on comparing irrational numbers based on square roots. Whereas this task is trivial if we just want to compare two line segments (e.g. by comparing the radicands), the problem starts when we want to compare two polygonal paths. Even assuming rational (or, after scaling, integer) coordinates, this problem translates into a question that is fundamental in computational geometry: Given two lists of integers, a1 ... and b1 ..., can we decide whether "∑ √ai ≥ ∑ √bi" in P? Put into words: Can we efficiently compare two sums of square roots over integers?
Continue reading