There is a folklore conjecture that $\operatorname{Ext}^2$ vanishes in the category of geometric $p$-adic Galois representations (i.e. representations that are unramified almost everywhere and de Rham at places dividing $p$) over a number field. See for example Conjecture II.3.2.2 of Fontaine--Perin-Riou's Autour des conjectures de Bloch et Kato. This is an avatar in Galois representations of Beilinson's conjecture that the category of mixed motives over a finitely-generated field $k$ has cohomological dimension at most the Kronecker dimension of $k$ (c.f. Remarks 4.12(c) in Uwe Jannsen, Motivic Sheaves and Filtrations on Chow Groups, Proceedings of Symposia in Pure Mathematics 55 (1994), Part 1, pp. 245–302, doi:10.1090/pspum/055.1, pdf).
What is the current state of the art on this conjecture? I was hoping that the experts here could piece together a complete picture of what's currently known.
Part 2 of Theorems A and B of
- Patrick B. Allen, Deformations of polarized automorphic Galois representations and adjoint Selmer groups, Duke Math. J. 165, no. 13 (2016), 2407-2460, doi:10.1215/00127094-3477342, arXiv:1411.7661,
building on earlier work by Flach and Wiles on modularity lifting, proves specific cases of $\operatorname{Ext}^2$ in the category of $p$-adic Galois representations unramified outside a finite set of places, but with no $p$-adic Hodge theory condition. However, as emphasized to me by David Loeffler, this does NOT prove the corresponding statement in the category of geometric Galois representations.
I would surmise that if there are results in this directions, they would be proven by Galois deformations or Iwasawa theory. In the latter case, I wonder if it might require computationally checking that a certain $p$-adic $L$-value does not vanish in order to prove the conjecture for a specific Galois representation.