why is centroid negative

My data are all non-negative. How can a centroid from kmeans have a negative value? My data are time series and I use most distance functions (to compare clusters). Each centroid is a time series, and should have C(t) >= min(data). Am I missing something?

[Image Analyst]

So you have a scatter plot with the "x" axis being time, t, and the "y" axis being C? And neither x nor y (t nor C(t)) have negative values. Is that correct?

[Harvey]

that is correct. I have t = 1:H and x >= 0, where x is N by H matrix. The centroid has C(t) < 0.

[Image Analyst]

Please supply some sample data that illustrates your situation.