Fruchterman-Rheingold layout - iteration decomposition dependence?

Mar 14, 2009 at 11:31 PM
I'm trying to improve my intuition for how the FR layout works.  Starting each time from blank X and Y coordinates, I create a chart in two ways, both times with the same strength coefficient (3.0):

1. Set 25 iterations, "read workbook", then "layout again", for a total of 50 iterations

2. Set at 50 iterations, click "read workbook"

The resulting layouts are different.  (The 50 iteration one is more compact than the 25+25.) Why?

Thanks for any insights

--- Pierre

>> Thanks, as always, to Tony, Marc et al. for a marvelous tool!
Mar 15, 2009 at 10:16 PM
Edited Mar 15, 2009 at 10:20 PM

It has to do with something Fruchterman and Reingold call a "temperature," which determines the maximum distance vertices can be displaced with each iteration of the algorithm.  The temperature is decreased with each iteration, which encourages the graph to stabilize over time, or "cool."

The initial temperature is proportional to the size of the graph pane, which is constant in both of your test cases, but the amount by which the temperature is decreased with each iteration is inversely proportional to the number of iterations.  Thus, the temperature sequences in your test cases differ --the temperature jumps back to its maximum value in the second step of your first case, but never jumps up in the second case-- and the vertices are displaced by different amounts.

-- Tony

Mar 16, 2009 at 1:59 AM

Ah, very helpful! Thanks, Tony

--- Pierre