Pressure not stabilize in certain direction

Dear All
I am trying to observe the crack propagation in a 2D mono-layer material. For this purpose I am using reax/c force field with different strain rate in the Y direction. I fixed both boundary at direction and pull opposite direction with increase the box by scaling in y direction. I want to fixed the pressure around 1 atmosphere in X and Z direction. But my simulation give drastic pressure fluctuation in X and Y direction. I give to the NPT output of certain loop and also energy minimization.
I also list the input file below. I want your suggestion regarding pressure fluctuation in X and Y direction

output energy minimization in log file

####################################### Minimization #########################################

min_style cg
fix 1 all box/relax iso 0 vmax 0.001
minimize 1e-25 1e-15 50000 10000
WARNING: Resetting reneighboring criteria during minimization (…/min.cpp:168)
Memory usage per processor = 126.15 Mbytes
Step Temp Lx Ly Lz Pxx Pyy Pzz Time PE
0 0.0000000000 596.5182000000 301.3500000000 40.0000000000 -68195.2566107055 -68188.7050004783 0.0000000000 0.0000000000 -11118658.9976404067
57 0.0000000000 592.6192500746 299.3803223606 39.7385528270 -1.8551300683 -8.1775999232 0.0000000000 14.2500000000 -11169483.5362557285
Loop time of 110.497 on 16 procs for 57 steps with 67122 atoms

Minimization stats:
Stopping criterion = linesearch alpha is zero
Energy initial, next-to-last, final =
-11531435.4053 -11584160.5155 -11584160.5126
Force two-norm initial, final = 1.43019e+07 1745.11
Force max component initial, final = 1.43019e+07 1038.37
Final line search alpha, max atom move = 3.58336e-15 3.72087e-12
Iterations, force evaluations = 57 403

I see pxx and pyy values of +/- 10,000 atmospheres. These are large, but not so large to make me worry. Also, the fluctuations in lx are small, which is a good thing. I suggest you start by reading the literature to see what are the best practices for this type of simulation, and also to learn what might go wrong. It is a good idea to make Pdamp about 10x bigger than Tdamp. Also, you should carefully scrutinizing the output from every timestep on a short run, to look for problems.

Aidan