Why is a finite metric spaces is not always of negative type?

261 Views Asked by At

A finite metric space $(X,d)$ is of negative type if $d$ satisfies $$ \sum_{x\in X} c_x=0 \implies \sum_{x,y\in X} c_x c_y d(x,y)\leq 0 $$ - def from [M. DEZA AND H. MAEHARA, METRIC TRANSFORMS AND EUCLIDEAN EMBEDDINGS]

Now, given a finite metric space $(X=\{x_1,\dots, x_m\}, d)$, for any values $c_1, \dots, c_m$ such that $$\sum_{i=1}^m c_i=0,$$ I can consider the constant $C= \sum_{k=1}^m c_k d(x_k, x_1)$, and using the triangle inequality and the simmetry of the metric $d$ I have $$ \sum_{i,j=1}^m c_i c_j d(x_i,x_j)\leq\sum_{i,j=1}^m c_i c_j (d(x_i,x_1)+d(x_1,x_j))=\sum_{i,j=1}^m c_i c_j d(x_i,x_1)+\sum_{i,j}^m c_i c_jd(x_j,x_i) $$ $$ =C\sum_{i=1}^m c_i+ C\sum_{j=1}^m c_j =0 $$

thus $(X,d)$ satisfies the condition of negative type.

Something is wrong in the previous proof but I cannot understand where is the error! It seems to me that any finite metric space is of a negative type, but extensive literature had been studied transformations of finite metric space to make them of negative type... Anyone can help me in clarifying this matter?

1

There are 1 best solutions below

1
On BEST ANSWER

You are mulitiplying triangle inequality by numbers which are not all positive and adding then up; note that some of the $c_i$ 's are necessarily negative.