gistfile1.rb

ruby-1.8.7-p249 > Geokit::Bounds.from_point_and_radius([53.91848,-122.776584], 4993)
 => #<Geokit::Bounds:0x1050e7580 @ne=#<Geokit::LatLng:0x1050e8e08 @lat=53.897792470202, @lng=-43.4942376451051>, @sw=#<Geokit::LatLng:0x1050e8e58 @lat=-18.265247529798, @lng=-202.058930354895>> 
ruby-1.8.7-p249 > Geokit::Bounds.from_point_and_radius([53.91848,-122.776584], 4997)
 => #<Geokit::Bounds:0x1050dee08 @ne=#<Geokit::LatLng:0x1050e0690 @lat=53.8399645290606, @lng=-43.4579700787395>, @sw=#<Geokit::LatLng:0x1050e06e0 @lat=-18.3230754709394, @lng=-202.09519792126>>

I guess that with such big distances, it's crossing the usual limits of -180/+180 for the longitude and -90/+90° for the longitude.

I think that Geokit does good with crossing the longitude limits, but I don't know about latitude.

Yep, I'm sure that's the problem.

As I pointed out the diameter either of those circles should be around 90 degrees in latitude, at 40 degrees north each degree of latitude is only 85 km, and at 54 north even less, so the diameter in longitude is greater than 120 degrees

Generalized spherical geometry for navigation has a singularity at each pole. Things get really wonky near the arctic circles, too. 54deg North is pretty "up there" I'd try shrinking your radius to 10km and then increase it again until you see things break.

Also, it depends on which algorithm GeoKit is using. There's a simplified formula which doesn't account for sign changes; it works fine for small ranges and is much faster; it just won't cross poles or the international dateline. In addition, like I said, these formulae start to break down as you get near the poles -- somewhere sine(theta) is approaching 0.