Freezing transition of the directed polymer in a 1+d random medium : location of the critical temperature and unusual critical properties

In dimension $d \geq 3$, the directed polymer in a random medium undergoes a phase transition between a free phase and a disorder dominated phase. For the latter, Fisher and Huse have proposed a droplet theory based on the scaling of the free energy fluctuations $\Delta F(l) \sim l^{\theta}$. On the other hand, in related growth models belonging to the KPZ universality class, Forrest and Tang have found that the height-height correlation function is logarithmic at the transition. For the directed polymer model at criticality, this translates into logarithmic free energy fluctuations $\Delta F_{T_c}(l) \sim (\ln l)^{\sigma}$ with $\sigma=1/2$. In this paper, we propose a droplet scaling analysis exactly at criticality based on this logarithmic scaling. Our main conclusion is that the typical correlation length $\xi(T)$ of the low temperature phase, diverges as $ \ln \xi(T) \sim (- \ln (T_c-T))^{1/\sigma} \sim (- \ln (T_c-T))^{2} $. Furthermore, the logarithmic dependence of $\Delta F_{T_c}(l)$ leads to the conclusion that the critical temperature $T_c$ actually coincides with the explicit upper bound $T_2$ derived by Derrida and coworkers, where $T_2$ corresponds to the temperature below which the ratio $\bar{Z_L^2}/(\bar{Z_L})^2$ diverges exponentially in $L$. Finally, since the Fisher-Huse droplet theory was initially introduced for the spin-glass phase, we briefly mention the similarities and differences with the directed polymer model. If one speculates that the free energy of droplet excitations for spin-glasses is also logarithmic at $T_c$, one obtains a logarithmic decay for the mean square correlation function at criticality $\bar{C^2(r)} \sim 1/(\ln r )^{\sigma}$.