Given the failure of all Dark Matter detection experiments so far, some have suggested that maybe classical understandings of Gravity (both Newtonian and the more complete General Relativity) are incomplete. The idea is that we have to tweak the equations a bit to explain the anomalous observed behavior at larger scales in the galaxy (and the local Universe). The tweaks are generally all thought to be part of a rethinking of our understanding of gravity and collected under the term of MOND (MOdified Newtonian Dynamics). 

There’s a new bit of evidence that such a change might be needed:

MOND: Crooked star clusters may be a sign that Einstein’s gravity is wrong | New Scientist:

Clusters of stars, which orbit the centre of their galaxy, typically look a bit like a two-armed pinwheel with opposing tails – though they don’t spin. Their tails are formed when stars bouncing around within the cluster begin to travel either slightly faster or slightly slower than the cluster as a whole. The leading tail in front of the cluster is made up of stars that are slightly closer to the galaxy’s centre, and the trailing one in the back is made of stars that are slightly further from the galactic centre and fall behind. In standard, or Newtonian, gravity, we would expect these two tails to be roughly equal – as the stars bounce around within the cluster, they should be equally likely to be thrown into either tail.

[…]

“It’s like there are two doors to escape the cluster, and the stars can only pass through the doors if they have the right direction and the right energy – otherwise they will just bounce around within the cluster,” says Kroupa. “In MOND, the front door is simply bigger.” Because of the way gravitational effects compound with one another in MOND, the forces pulling stars towards the centre of the galaxy, and therefore towards the leading tail, are stronger than in Newtonian gravity.

This is just a data point. It doesn’t overthrow the accepted theory on its own, but it’s sure interesting. General Relativity works extraordinarily well in almost all situations, and has passed test after test to the limits of our ability to measure.

If something different is really going on, it’s not going to be a simple change to a model…