What is geometrically the Pontryagin class?

What does the Pontryagin class detects or is an obstruction to? Please avoid any answer using that it's the even Chern class of the complexified bundle or any interpretation that relies on the complexified bundle.

As related question might be the following: when one defines the obstruction classes on a rank $4$ vector bundle (and if the first three obstruction classes do vanish) then the fourth obstruction class can be decomposed as the Euler class and the first Pontryagin class (as $\pi_3(SO_4) \simeq \mathbb{Z} \oplus \mathbb{Z}$). Is there a geometric description of a system of generators in $\pi_3(SO_4)$ which is associated to these classes?

EDIT: deleted "For example, why does the first Pontryagin class distinguishes the (tangent bundles of the) exotic $4$-spheres?" as it is wrong, see Liviu's answer below.

-
A related question (which I don't know the answer to): What was Pontryagin's motivation for introducing theses classes? (And if it wasn't him, who did introduce them, why?) – Charles Rezk Dec 22 at 20:01
$SO(4)$ is double covered by $SU(2)\times SU(2)$ and since $SU(2)=S^3$, $\pi_3(SO(4))=\pi_3(S^3)\times \pi_3S^3=Z\times Z$. – Paul Dec 22 at 23:19
@Charles: I would also like to hear about this if you ever get the answer. @Paul: thanks, but I was more looking to learn what is the system of generator $\langle \alpha, \beta \rangle$ of $\mathbb{Z} \times \mathbb{Z}$ so that given an element, if one writes it down as $a \alpha + b \beta$ then $a$ would be associated to the Euler class and $b$ to the Pontryagin class – Antoine Dec 23 at 10:05
Perhaps you are already aware of this, but generally one obtains interesting invariants as polynomials in the pontryagin classes rather than by looking at the pontryagin classes themselves - see the A-hat genus or the hirzebruch L-class, for instance. – Paul Siegel Dec 23 at 15:30

Some fractional Pontrjagin classes are obstructions to higher analogues of orientations/spin structures.

For example, a spin vector bundle $E \longrightarrow X$ admits a string structure if $\frac{1}{2}p_1(E) = 0$. In other words, a spin structure on $E$ determines a class $\lambda = \frac{1}{2} p_1(E) \in H^4(X; \mathbb{Z})$ such that $2\lambda = p_1(E)$, and this fractional first Pontrjagin class $\lambda$ is the obstruction to the existence of a string structure on $E$.

Similarly, if we go to the next nontrivial step on the Whitehead tower, we can try to define a so-called fivebrane structure on a string vector bundle $E \longrightarrow X$. In this case, the obstruction to the string vector bundle $E \longrightarrow X$ admitting a fivebrane structure is the fractional second Pontrjagin class $\frac{1}{6}p_2(E)$.

-
It's maybe worth noting that that pattern doesn't continue step-by-step. Moving from $B\mathrm{Fivebrane}$ to the next stage in the Whitehead tower is not described by Pontryagin classes. This isn't a very exciting thing to point out, though; for almost identical reasons, lifting unoriented real vector bundles to oriented ones and then to spin ones is also not controlled by Pontryagin classes. – Eric Peterson Dec 22 at 22:06

There is a paper by Paul Bressler:

The ﬁrst Pontryagin class

http://arxiv.org/abs/math/0509563

According to him:

We give a natural obstruction theoretic interpretation to the ﬁrst Pontryagin class in terms of Courant algebroids. ........

Thus, (A,h , i) admits a (globally deﬁned) Courant extension if and only if the the Pontryagin class of (A,h , i) vanishes.

More generally, the ﬁrst Pontryagin class with values as above may be associated to a transitive Lie algebroid (see A.1), say, A, together with an invariant symmetric pairing h , i on the kernel of the anchor map and will be denoted Π(A,h , i). 1 The Pontryagin class of a principal bundle is deﬁned as the Pontryagin class of the Atiyah algebra of the bundle.

-
 thanks, I'll try to look through this... – Antoine Dec 23 at 9:50

I don't think that $p_1$ distingushes the tangent bundles of exotic $4$-spheres (if any). On an oriented smooth $4$-manifold $M$ Hirzebruch signature formula states that

$${\rm sign}(M)=\frac{1}{3}\int_M p_1(TM).$$

The signature of any homology $4$-sphere is zero since there is no homology in the $4$-th dimension.

1. There is one stupid way in which $p_1$ describes an obstruction, because $p_1$ is the $2$-nd Chern class of the complexification, and Chern classes have obstruction-theoretic descriptions.

2. The first Pontryagin class of a $4$-manifold $M$ appears in a nice integral formula of MacPherson and it involves the singularities of generic maps $M\to \mathbb{R}^4$. (I do not remember the reference at this moment.)

-
 My mistake about the 4-spheres: I thought it was in a paper of Briekorn (which is about 7-spheres and does not mention Pontryagin class so I was way off)... – Antoine Dec 23 at 9:59