1: \begin{abstract}

2: The failure of observing the $e^+ e^- \to J/\psi+J/\psi$ events at $B$ factories to date

3: is often attributed to the significant negative order-$\alpha_s$ correction.

4: In this work we compute the ${\cal O}(\alpha^2_s)$ correction to this process for the first time.

5: The magnitude of the next-to-next-to-leading order (NNLO) perturbative correction is substantially negative

6: so that the standard NRQCD prediction would suffer from an unphysical, negative cross section.

7: This dilemma may be traced in the fact that the bulk contribution of the fixed-order radiative corrections stems from the

8: perturbative corrections to the $J/\psi$ decay constant.

9: We thus implement an improved NRQCD factorization framework, by decomposing the amplitude into the photon-fragmentation piece and

10: the non-fragmentation piece. With the measured $J/\psi$ decay constant as input, which amounts to resumming a specific class of radiative and

11: relativistic corrections to all orders, the fragmentation-induced production rate can be predicted accurately and serves a benchmark prediction.

12: The non-fragmentation type of the amplitude is then computed through NNLO in $\alpha_s$ and at lowest order in velocity.

13: Both the ${\cal O}(\alpha_s)$ and ${\cal O}(\alpha^2_s)$ corrections in the interference term become positive and exhibit a decent convergence behavior.

14: Our finest prediction is $\sigma(e^+ e^- \to J/\psi+J/\psi)=  2.13^{+0.30}_{-0.06}$ fb  at $\sqrt{s}=10.58$ GeV.  With

15: the projected integrated luminosity of 50 ${\rm ab}^{-1}$, the prospect to observe this exclusive

16: process at \texttt{Belle} 2 experiment appears to be bright.

17: \end{abstract}

18: