According to my lecture notes, these three definitions of flat modules are equivalent:
a) For every exact sequence of $R$-modules $N’ \rightarrow N \rightarrow N’’$, the sequence $N’ \otimes_R M \rightarrow N \otimes_R M \rightarrow N’’ \otimes_R M $ is exact.
b) For every short exact sequence of $R$-modules $0 \rightarrow N’ \rightarrow N \rightarrow N’’ \rightarrow 0$, the sequence $0 \rightarrow N’ \otimes_R M \rightarrow N \otimes_R M \rightarrow N’’ \otimes_R M \rightarrow 0$ is exact.
c) For every exact sequence of $R$-modules $0 \rightarrow N’ \rightarrow N \rightarrow N’’$, the sequence $0 \rightarrow N’ \otimes_R M \rightarrow N \otimes_R M \rightarrow N’’ \otimes_R M $ is exact.
I can see that c) implies b) because of the right-exactness of the tensor product, but I can’t see why b) implies a).
I thought about “splitting” the exact sequence $N’ \rightarrow N \rightarrow N’’$ into short exact sequences, for instance $0 \rightarrow \ker(\alpha) \rightarrow N’ \rightarrow \mbox{im}(\alpha) \rightarrow 0$ and $0 \rightarrow \ker(\beta) \rightarrow N \rightarrow \mbox{im}(\beta) \rightarrow 0$ and trying to apply b) to these, then to patch them together to obtain a) using that $\ker(\beta)=\mbox{im}(\alpha)$, but to no avail (here $\alpha:N’ \rightarrow N$ and $\beta: N \rightarrow N’’$).
I realize this has been asked before on this site, but this was in 2012 and I don’t understand why the answer solves the problem (is there a reason why $\mbox{coker} (\alpha) \cong N’’$?).
Also, I don’t see why a) implies c).