About the difference between fcm and compute reduce sum

LAMMPS
1

Hi all,

I have used and compared the following commands in the simulation:
(1) variable fx_group equal fcm(group,x)
(2) compute fx_group group reduce sum fx
(3) fix f_group group setforce NULL NULL NULL
Based on the explanation in the manual, I think these commands will obtain the same force result on the group.
Through the output in the thermo info, I found that the results of commands 2 and 3 are exactly the same. However, there is a very small difference between commands 1 and 2; the difference is in the order of 1e-14.
I have tested this in the in.friction example file. This is the test script

# 2d friction simulation

dimension	2
boundary	p s p

atom_style	atomic
neighbor	0.3 bin
neigh_modify	delay 5

# create geometry

lattice		hex 0.9
region		box block 0 50 0 22 -0.25 0.25
create_box	4 box

mass		1 1.0
mass		2 1.0
mass		3 1.0
mass		4 1.0

# atom regions

region          lo-fixed block INF INF INF 1.1 INF INF
region          lo-slab block INF INF INF 7 INF INF
region          above-lo block INF INF INF 7 INF INF side out
region          hi-fixed block INF INF 20.9 INF INF INF
region          hi-slab block INF INF 15 INF INF INF
region          below-hi block INF INF 15 INF INF INF side out
region		lo-asperity sphere 32 7 0 8
region		hi-asperity sphere 18 15 0 8
region		lo-half-sphere intersect 2 lo-asperity above-lo
region		hi-half-sphere intersect 2 hi-asperity below-hi

# create 2 surfaces with asperities

create_atoms	1 region lo-slab
create_atoms	1 region hi-slab
create_atoms	2 region lo-half-sphere
create_atoms	3 region hi-half-sphere

# LJ potentials

pair_style	lj/cut 2.5
pair_coeff	* * 1.0 1.0 2.5

# define groups

group	        lo region lo-slab
group		lo type 2
group		hi region hi-slab
group		hi type 3
group		lo-fixed region lo-fixed
group		hi-fixed region hi-fixed
group		boundary union lo-fixed hi-fixed
group		mobile subtract all boundary

set		group lo-fixed type 4
set		group hi-fixed type 4

# initial velocities

compute         new mobile temp/partial 0 1 0
velocity	mobile create 0.1 482748 temp new
velocity	hi set 1.0 0.0 0.0 sum yes

# Force analysis
fix        f_hi           hi-fixed      setforce  NULL NULL NULL
variable   fx_hi	  equal   fcm(hi-fixed,x)
compute    fx_hi	  hi-fixed      reduce sum fx
variable   test1          equal   f_f_hi[1]-v_fx_hi
variable   test2          equal   f_f_hi[1]-c_fx_hi


# fixes

fix		1 all nve
fix		2 boundary setforce NULL NULL NULL
fix		3 mobile temp/rescale 200 0.1 0.1 0.02 1.0
fix_modify	3 temp new
fix             4 all enforce2d

# Run

timestep	0.0025
thermo		1000
thermo_style    custom  time temp v_test1 v_test2
thermo_modify	temp new

dump		1 all atom 500 dump.friction

#dump		2 all image 500 image.*.jpg type type &
#		zoom 1.6 adiam 1.5
#dump_modify	2 pad 5

#dump		3 all movie 500 movie.mpg type type &
#		zoom 1.6 adiam 1.5
#dump_modify	3 pad 5

run		20000

Could you please help me understand the reason for this difference? Is this a numerical error? Or do I not understand the function of these commands?

I really appreciate your kind help and time.

Best regards,
Shusen Guo

2

This is within the precision of double precision floating point values. The computation in compute reduce is a bit more complex than which is done in the other two options, which can explain a small different that is within the floating point epsilon due to slightly different operations and ordering.

For that you would have to look at the source code in fix_setforce.cpp, compute_reduce.cpp, variable.cpp and group.cpp

3

Hi akohlmey,

I appreciate your kind help, this is really helpful!
Could you please help me with which option is more computationally efficient?

Best regards,
Shusen Guo