To Do List (current)

**Jan 2017**

* **Dries**: - [ ] calculate constrained CD protocols numerically * **Marin**: - [ ] MF spin glass theory: from a paramagnet there's a dynamical transition in time used for GD and a static transition from exponentially many minima to polynomially many (same as checking the entropy, see below). - [ ] Look at entropy - [ ] define correlation time (number of consecutive time steps which matter for obtaining unit fidelity) vs T. In MB system this might always be as big as the total ramp time T. - [ ] finish plots for 2LS part of paper - [x] smear out protocols at a fixed $\delta t$ and re-calculate $q_{EA}$ * **Alex**: - [ ] Run SA at finite T for 2LS and get fidelities - [ ] Check the effect of increasing hmax on the $T_{min}$ transition (kink) - [x] get DOS by random sampling - [x] get all 400 realisations for all T and dt (need for entropy) - [x] run $dt=0.01, 0.005, 0.0025, 0.0001$ to get the odd points - [x] Implement fast fidelity evaluations by precomputing and storing evolution matrices. - [x] Adjust code for SA (small fixes). - [x] calculate $q_{EA}$ for dt = 0.005, 0.02, 0.05 and $T \in [0.0, 3.0]$ - [x] Run scc with dt variable but nstep=$100,150,200$ - [x] Check PCA plots. - [x] Check what happens when you reduce $dt$. Try $dt=0.01,0.025$. - [x] Fix the number of potential bangs by adjusting dt. - [x] Run jobs on the scc - setup working environments. - [x] Check EA OP for continuous protocol - [x] Note on EA order parameter, email Marin, Pankaj - [x] Run simulations (SGA) : Fidelity vs evolution time. Fidelity vs. computation time. - [x] Do the F vs total time T curve - [x] Save the optimal protocols and the inst. fidelity vs iteration number within every run - [x] Run codes for following specifications and compute the following quantities: - actions=[-0.8,0.,0.8] - continuous protocol: +- [0,0.1,0.2,0.5,1.0,2.0,4.0,8.0] - psi_i=GS(hx=-1), psi_f=GS(hx=+1) - hx\in[-4,4] [the initial field is pre-quenched to -4.0] - Time slices: np.linspace(0.25,2.5,10). Running 2E4 episodes (Marin) - [x] Compute EA order parameter - [x] Perform t-SNE plots of obtained bang-bang spin configurations (could be Potts spin too)

• MB part:

• Check MB protocol for stability

• further studies:

• idea: train at positive temperature, then flip sign of temperature and initiate with pre-trained Q
• change direction of sweeping (is it possible?)
• write Q-learning based on ADAMS
• figure out which temperature is given by the DOS in reward space