How Does Dark Matter Form Lumps? As far as we know, the particles of dark matter can interact with each other only by gravitation.  No electromagnetics, no weak force, no strong force.  So, let's suppose a local slight concentration of dark matter comes about by chance motions and begins to gravitate.  The particles would fall "inward" towards the center of the concentration. However, with no interaction to dissipate angular momentum, they would just orbit the center of the concentration and fly right back out to the vicinity of where they started resulting in no increase in density.  Random motions would eventually wipe out the slight local concentration and we are left with a uniform distribution again.

How does dark matter form lumps?

 A: In fact, three and more body purely gravitational interactions can form clusters and clumps by concentrating some particles and expelling others, as mentioned in a previous answer.  Astronomers have discovered that this seems to happen quite rapidly. They call it "violent relaxation".  Google "violent relaxation" for more info.
A: My problem is with the time scale of the phenomenon.  It has been proposed that the universal network of filaments and concentrated lumps of dark matter is the framework upon which ordinary matter condensed to form our present day clusters of galaxies.  If this is so, it would seem reasonable that there be some rapid mechanism for its formation early in the development of the universe.  With only gravity to draw it together and mechanisms like the three body interaction expelling one of the bodies to permit coalescence into higher density lumps, how did the dark matter network manage to form first?  Ordinary matter has the same gravitational means of aggregating plus the electromagnetic interactions for reducing angular momentum.  Why didn't the ordinary matter condense first?
A: I suggest dark matter loses energy by proxy though gravitational interactions with ordinary matter that is losing energy through radiative processes.  As ordinary matter loses energy  by radiation allowing it to clump gravitationally, dark matter clumps along with it by losing energy to the cooling ordinary matter. 
A: Cold dark matter is usually assumed to interact only under gravity and computing simulations are done with N particles to see how they would arrange themselves on the scales of galaxies, or globular clusters. In order to clump under gravity, the dark matter particles need to lose angular momentum, this is done from 3 body interactions, where a pair of particles 'kicks' another particle to have a higher angular momentum while the pair lose some of their angular momentum and get closer in orbit to the center of mass. Dark Matter researchers then find the profile of the dark matter from the simulations, usually finding a cusp like center (1/r)^n density profile (n<-1, I've seen n=1.7 often, but this varies with the detail of the simulations). Dark Matter (especially the supersymmetric LSP type), also might self annihilate, from point-like weak forces when two particle overlap, this is a rare process but which leads to observable high energy gamma rays, which would have a (1/r)^2n like profile. Dark Matter researchers are currently divided on whether such gamma rays have been observed (e.g. by the Fermi satellite and H.E.S.S. telescopes), and also worry whether the dark matter HALO around galaxies is well described by (1/r)^n scale profiles. If researchers don't see something like a (1/r)^-2.5 gamma ray pattern from the galactic center or something close, they may need to find alternative models for dark matter. 
A: The prevailing theory of dark matter is the Cold Dark Matter (CDM) hypothesis. This hypothesis is favored because it is assumed that the dark matter particles are non-relativistic - i.e. slow moving. Because they are slow moving, they can essentially orbit in and around the original small density fluctuations, making these small density fluctuatuions stable. These small density fluctuations can clump into denser clumps due to three body gravitational interactions. A three body interaction of small clumps can result in one of the clumps being ejected at a higher speed while the other two clumps slow down and become more gravitationally bound.
However, baryonic matter can clump more effectively than CDM since the electromagnetic interractions allow baryonic matter to cool more effectively than the CDM clumps.  That is why the prevailing theory is that the DM forms halos that are more distended than the clumps of baryonic matter.  Thus a visible galaxy will have a more extended DM halo that extends far beyond the visible stars of the galaxy. The DM halo will also be more spherical than the flattened galactic disk.
It is true that most DM models assume the DM particles do have weak interactions, but these interactions aren't required by the CDM model. However, these weak interactions are required if any of the dark matter detection experiments are to be successful.
[Note: After more research, I discovered that my original answer was wrong. I now think this answer is correct. Sorry about that.]
