Do smooth manifolds admit linear atlases? There is a theorem of Whitney showing that a smooth manifold can be endowed with a compatible real-analytic atlas (later, it was proven that this analytic structure is essentially unique).
I am curious, how much stronger structures can be put on a smooth manifold in a compatible way? Most importantly, is it possible to find an atlas such that its transition maps be elements of $GL(n)$ ($n$ being the dimension of the manifold)?
The only related thing that I have found is the concept of piecewise-linear manifold, which seems not to be what I am looking for.
 A: In dimension 3, radiant affine manifolds have been classified by Choi. To quote from the AMS Math Review: 

The main result is a decomposition theorem, which says that such a manifold admits a decomposition along finitely many disjoint totally geodesic tori or Klein bottles into two kinds of pieces. The first kind are convex radiant affine 3-manifolds, i.e. those for which the universal cover is real projectively homeomorphic to a convex domain in affine space. The second kind of piece is the suspension of a real projective surface of a special kind by a real projective automorphism. If Σ is a real projective (n−1)-manifold and ϕ is a real projective automorphism of Σ, there is a natural radial affine structure on the mapping torus of the automorphism ϕ, which is a quotient by a cyclic group of a component of the complement of the zero section of the tautological line bundle over Σ coming from its projective structure. 
     If M is a closed manifold with a radiant affine structure, there is a natural developing map from the universal cover M˜ toRn−{0}, unique up to composition with an element of GL(n,R). The radial vector field X=∑ixi∂/∂xi on Rn−{0} pulls back by the developing map to a vector field on M˜ which by naturality covers a vector field on M. This vector field is known as the radial vector field on M. Carrière asked whether every compact radiant affine 3-manifold admits a total cross section to the radial flow. Together with some results of Barbot and the author, contained in an appendix, the decomposition theorem answers the Carrière conjecture in the positive. Together with some basic results in 3-manifold topology, this implies that every compact radiant affine 3-manifold with empty or totally geodesic boundary is homeomorphic to a Seifert space with Euler number zero, or is finitely covered by a surface bundle over a circle with fiber homeomorphic to a compact surface of Euler characteristic zero. 

A: An $n$-dimensional manifold endowed with an atlas whose transition function are in $Gl(n,R)$ is called a radiant affine manifold.
An affine manifold $M$ is defined by an atlas whose coordinate changes are in $Aff(R^n)$, the universal cover $\hat M$ inherits an affine structure defined by a local diffeomorphism 
$D_M:\hat M\rightarrow R^n$ which induces a representation 
$h_M:\pi_1(M)\rightarrow Aff(R^n)$ defined by $h_M(g)=(L(h_M)(g),a_g)$.
Here $L(h_M)$ is the linear part of $h_M$. The representation $L(h_M)$ is the holonomy of a connection defined on $M$ whose curvature and torsion vanish. 
The correspondence $g\rightarrow a_g$ is a $1$-cocycle for the linear holonomy $L(h_M)$, if its class is zero then the structure is radiant and the structure is defined by coordinates change which take their values in $Gl(n,R)$.
