Question about "compute chunk/msd"

Hi everyone,
I am trying to compute diffusion coefficient of atoms in a specific sphere by compute chunk/msd, but the results seemed unreasonable. I plotted the msd versus steps in each spheres, but the plots show no relationship between msd and timestep (Figure 1). However, I tried compute msd without sphere, a linear relationship can be seen (Figure 2). I would like to know that did I make some mistakes in using compute msd/chunk? Thank you very much for replying.

Here is my script:

units		    				real
boundary						p p p
atom_style	    		charge
atom_modify	    		map array sort 5000 2
read_data						matrix.dat

#----------------------------Simulation settings-----------------------
# 1 is O	2 is Si	3 is Al	4 is Ca
pair_style					hybrid/overlay coul/dsf 0.25 8.0 table linear 8000
pair_coeff					* * coul/dsf
pair_coeff					1 1 table Pedone_CAS.table OO
pair_coeff					1 2 table Pedone_CAS.table SiO
pair_coeff					1 3 table Pedone_CAS.table AlO
pair_coeff					1 4 table Pedone_CAS.table CaO

neighbor						2.0 bin
neigh_modify				delay 0 every 5 check yes
thermo	    				1000
thermo_style				custom step atoms temp pe lx pxx ly pyy lz pzz density
timestep						2

velocity						all create 300 1234 dist gaussian
fix	        				1 all npt temp 300 1800 200 iso 1.01325 1.01325 2000
run		        			150000
unfix	    					1

group								O type 1
group								Si type 2
group								Al type 3
group								Ca type 4

variable						cx equal "(xhi+xlo)/2"
variable						cy equal "(yhi+ylo)/2"
variable						cz equal "(zhi+zlo)/2"

compute							bins_O O chunk/atom bin/sphere ${cx} ${cy} ${cz} 10 50 200
compute							bins_Si Si chunk/atom bin/sphere ${cx} ${cy} ${cz} 10 50 200
compute							bins_Al Al chunk/atom bin/sphere ${cx} ${cy} ${cz} 10 50 200
compute							bins_Ca Ca chunk/atom bin/sphere ${cx} ${cy} ${cz} 10 50 200

compute							msd_O all msd/chunk bins_O
compute							msd_Si all msd/chunk bins_Si
compute							msd_Al all msd/chunk bins_Al
compute							msd_Ca all msd/chunk bins_Ca

fix									1 all nve
fix 								2 all ave/time 1 1 1 c_msd_O[4] c_msd_Si[4] c_msd_Al[4] c_msd_Ca[4] file msd_chunk_matrix.txt mode vector
run									250000
unfix								1
fix									3 all nvt temp 1800 1800 200
run									250000

Figure1 compute msd/chunk

Figure2 compute msd

If you want to compute diffusivity via MSD, then - by definition - you need to follow particles for a long time because only in the limit of infinite time, does the Einstein relation apply. But that concept goes contrary to subdividing your system into (small) chunks. The chunk mechanism is much more suited to study instantaneous properties like density or kinetic energy or stress.