[BUG] MPI errors when running NEB with multiple cores per replica
jcappola opened this issue · 5 comments
Summary
Running NEB with multiple cores per replica, such as
mpirun -np 8 lmp -partition 4x2 -in in.neb.sivac leads to a few different MPI-related crashes.
LAMMPS Version and Platform
LAMMPS (2 Aug 2023) - Update 2 running on Ubuntu 20.04.6 LTS compiled with Open-MPI 4.0.3 including the MANYBODY and REPLICA package
Expected Behavior
The free-end NEB end commands should (ideally) all allow for multiple cores per replica to be run. At the moment, we are limited to a single core per replica for end last/efirst and end last/efirst/middle.
Actual Behavior
When running NEB with multiple cores per replica, e.g.,
mpirun -np 8 lmp -partition 4x2 -in in.neb.sivac
LAMMPS will return an error if no verbosity argument is provided to the neb command:
*** An error occurred in MPI_Allgather
*** reported by process [2652307457,4]
*** on communicator MPI COMMUNICATOR 4 SPLIT FROM 0
*** MPI_ERR_TRUNCATE: message truncated
*** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
*** and potentially your MPI job)
This issue is immediately fixed by adding any verbosity argument to the neb command.
Separately, when trying to use free-end NEB (particularly only end last/efirst and end last/efirst/middle) with multiple cores per replica will return a different MPI error:
*** An error occurred in MPI_Bcast
*** reported by process [2639265793,2]
*** on communicator MPI_COMM_WORLD
*** MPI_ERR_COMM: invalid communicator
*** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
*** and potentially your MPI job)
which completely prevents the use of multiple cores per replica for more complicated free-end NEB calculations. Note that this issue does not occur for end first or end last. This has been an constant issue on multiple platforms and installations.
Steps to Reproduce
After a clean install of LAMMPS (2 Aug 2023) - Update 2 including the MANYBODY and REPLICA package, you can run the example/neb case in.neb.sivac with the correct partition command to replicate these issues.
Further Information, Files, and Links
It seems that the behavior for the first error is fixed by just adding the verbosity keyword to the neb command, although it is not clear to me as to why. There is only a single MPI_Allgather that a NEB script running with the default print_mode should encounter which is on line 642 of neb.cpp.
The second error seems to be fixed by changing the communicator in line 302 of fix_neb.cpp from:
if (me == 0) MPI_Bcast(&vIni, 1, MPI_DOUBLE, 0, rootworld);
to:
if (me == 0) MPI_Bcast(&vIni, 1, MPI_DOUBLE, 0, uworld);
although this is my first time poking around the LAMMPS source code so I don't know if this is the correct/desired fix.
There is no "end" keyword in the "neb" command: https://docs.lammps.org/neb.html
So please provide the exact input files that cause the issue and the corresponding log or screen output files for replica 0.
I have no problems with the provided example using either OpenMPI 4.1.5 or MPICH 4.1.2 with a 4x2 partition.
I've tested the last stable and the current development version.
Hi @akohlmey,
I am aware that the "end" keyword is under the "fix neb" command, sorry for any confusion.
On a clean install of LAMMPS (2 Aug 2023) - Update 2, I run the "example/neb/in.neb.sivac" case with:
mpirun -np 8 lmp -partition 4x2 -in in.neb.sivac
where the NEB region of that original example script looks like this:
fix 1 all neb 1.0
thermo 100
# run NEB for 2000 steps or to force tolerance
timestep 0.01
min_style quickmin
neb 0.0 0.01 100 100 10 final final.sivac
and I get the following error print to the screen:
[jcappola-Precision-Tower-7910:07055] *** An error occurred in MPI_Allgather
[jcappola-Precision-Tower-7910:07055] *** reported by process [4100849665,4]
[jcappola-Precision-Tower-7910:07055] *** on communicator MPI COMMUNICATOR 4 SPLIT FROM 0
[jcappola-Precision-Tower-7910:07055] *** MPI_ERR_TRUNCATE: message truncated
[jcappola-Precision-Tower-7910:07055] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[jcappola-Precision-Tower-7910:07055] *** and potentially your MPI job)
All of the screen.* files are blank and the log.lammps.* file for partition 0 looks like this:
LAMMPS (2 Aug 2023)
Processor partition = 0
OMP_NUM_THREADS environment is not set. Defaulting to 1 thread. (src/comm.cpp:98)
using 1 OpenMP thread(s) per MPI task
# NEB simulation of vacancy hopping in silicon crystal
units metal
atom_style atomic
atom_modify map array
boundary p p p
atom_modify sort 0 0.0
# coordination number cutoff
variable r equal 2.835
# diamond unit cell
variable a equal 5.431
lattice custom $a a1 1.0 0.0 0.0 a2 0.0 1.0 0.0 a3 0.0 0.0 1.0 basis 0.0 0.0 0.0 basis 0.0 0.5 0.5 basis 0.5 0.0 0.5 basis 0.5 0.5 0.0 basis 0.25 0.25 0.25 basis 0.25 0.75 0.75 basis 0.75 0.25 0.75 basis 0.75 0.75 0.25
lattice custom 5.431 a1 1.0 0.0 0.0 a2 0.0 1.0 0.0 a3 0.0 0.0 1.0 basis 0.0 0.0 0.0 basis 0.0 0.5 0.5 basis 0.5 0.0 0.5 basis 0.5 0.5 0.0 basis 0.25 0.25 0.25 basis 0.25 0.75 0.75 basis 0.75 0.25 0.75 basis 0.75 0.75 0.25
Lattice spacing in x,y,z = 5.431 5.431 5.431
region myreg block 0 4 0 4 0 4
#create_box 1 myreg
#create_atoms 1 region myreg
#mass 1 28.06
#write_data initial.sivac
read_data initial.sivac
Reading data file ...
orthogonal box = (0 0 0) to (21.724 21.724 21.724)
1 by 1 by 2 MPI processor grid
reading atoms ...
512 atoms
reading velocities ...
512 velocities
read_data CPU = 0.010 seconds
# make a vacancy
group Si type 1
512 atoms in group Si
group del id 300
1 atoms in group del
delete_atoms group del compress no
Deleted 1 atoms, new total = 511
group vacneigh id 174 175 301 304 306 331 337
7 atoms in group vacneigh
# choose potential
pair_style sw
pair_coeff * * Si.sw Si
Reading sw potential file Si.sw with DATE: 2007-06-11
# set up neb run
variable u uloop 20
# initial minimization to relax vacancy
displace_atoms all random 0.1 0.1 0.1 123456
Displacing atoms ...
minimize 1.0e-6 1.0e-4 1000 10000
Neighbor list info ...
update: every = 1 steps, delay = 0 steps, check = yes
max neighbors/atom: 2000, page size: 100000
master list distance cutoff = 5.77118
ghost atom cutoff = 5.77118
binsize = 2.88559, bins = 8 8 8
1 neighbor lists, perpetual/occasional/extra = 1 0 0
(1) pair sw, perpetual
attributes: full, newton on
pair build: full/bin/atomonly
stencil: full/bin/3d
bin: standard
Per MPI rank memory allocation (min/avg/max) = 4.113 | 4.114 | 4.114 Mbytes
Step Temp E_pair E_mol TotEng Press
0 0 -593.40319 0 -593.40319 355294.02
33 0 -2213.3343 0 -2213.3343 -3383.2606
Loop time of 0.042342 on 2 procs for 33 steps with 511 atoms
97.7% CPU use with 2 MPI tasks x 1 OpenMP threads
Minimization stats:
Stopping criterion = energy tolerance
Energy initial, next-to-last, final =
-593.403188091472 -2213.33209897182 -2213.33426537417
Force two-norm initial, final = 1101.8254 0.16683659
Force max component initial, final = 334.49264 0.014961353
Final line search alpha, max atom move = 1 0.014961353
Iterations, force evaluations = 33 44
MPI task timing breakdown:
Section | min time | avg time | max time |%varavg| %total
---------------------------------------------------------------
Pair | 0.037702 | 0.037902 | 0.038102 | 0.1 | 89.51
Neigh | 0.00096726 | 0.00098613 | 0.001005 | 0.0 | 2.33
Comm | 0.0021386 | 0.0023585 | 0.0025785 | 0.5 | 5.57
Output | 0 | 0 | 0 | 0.0 | 0.00
Modify | 0 | 0 | 0 | 0.0 | 0.00
Other | | 0.001095 | | | 2.59
Nlocal: 255.5 ave 258 max 253 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Nghost: 1080.5 ave 1083 max 1078 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Neighs: 0 ave 0 max 0 min
Histogram: 2 0 0 0 0 0 0 0 0 0
FullNghs: 8689 ave 8777 max 8601 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Total # of neighbors = 17378
Ave neighs/atom = 34.007828
Neighbor list builds = 1
Dangerous builds = 0
All the other log.lammps.* files are identical except for the timings on the CPUs. Note that 2 MPI tasks were reported on each partition here.
When I then make the modification to the "in.neb.sivac" input script as below:
fix 1 all neb 1.0
thermo 100
# run NEB for 2000 steps or to force tolerance
timestep 0.01
min_style quickmin
neb 0.0 0.01 100 100 10 final final.sivac verbosity default
I get the expected output to the screen:
LAMMPS (2 Aug 2023)
Running on 4 partitions of processors
Reading NEB coordinate file(s) ...
Setting up regular NEB ...
Step MaxReplicaForce MaxAtomForce GradV0 GradV1 GradVc EBF EBR RDT RD1 PE1 RD2 PE2 RD3 PE3 RD4 PE4
0 7.5525391 2.671788 0.16683659 7.5525391 7.5525391 1.5383951 0 1.6207355 0 -2213.3343 0.33333333 -2212.7428 0.66666667 -2212.2247 1 -2211.7959
10 0.24005275 0.0013324036 0.036483049 0.24005275 0.68351722 0.42916118 0.41794425 1.6989349 0 -2213.3365 0.32909183 -2212.9587 0.65386736 -2212.9073 1 -2213.3253
20 0.07940898 0.00026889621 0.024706844 0.07940898 0.71637784 0.41387872 0.41157886 1.7343662 0 -2213.3369 0.32478734 -2212.9621 0.65348766 -2212.923 1 -2213.3346
30 0.094973708 6.9942581e-05 0.015145947 0.035267404 0.7535772 0.40072717 0.40024605 1.7504612 0 -2213.3372 0.32705584 -2212.9584 0.65894506 -2212.9365 1 -2213.3367
40 0.027727472 1.9827556e-05 0.011618173 0.022562656 0.76133752 0.39614635 0.39591731 1.7547519 0 -2213.3373 0.32873163 -2212.9562 0.66124255 -2212.9411 1 -2213.337
50 0.01942934 9.0662902e-06 0.0087135565 0.015391975 0.7695268 0.39274846 0.3926388 1.7578616 0 -2213.3373 0.33022595 -2212.9543 0.66307279 -2212.9446 1 -2213.3372
60 0.019056184 2.6462904e-06 0.0053426943 0.0086167383 0.77759662 0.38936868 0.38933371 1.7610433 0 -2213.3374 0.33187545 -2212.9523 0.66497614 -2212.948 1 -2213.3373
63 0.0097002883 1.6300083e-06 0.0047744861 0.0076067229 0.77865612 0.38888517 0.38885789 1.7615322 0 -2213.3374 0.3321224 -2212.952 0.66525533 -2212.9485 1 -2213.3373
Setting up climbing ...
Climbing replica = 3
Step MaxReplicaForce MaxAtomForce GradV0 GradV1 GradVc EBF EBR RDT RD1 PE1 RD2 PE2 RD3 PE3 RD4 PE4
63 0.77865612 0.09663051 0.0047744861 0.0076067229 0.77865612 0.38888517 0.38885789 1.7615322 0 -2213.3374 0.3321224 -2212.952 0.66525533 -2212.9485 1 -2213.3373
73 0.098996478 0.0011347237 0.0027942177 0.0042838869 0.038660264 0.51024721 0.51023862 1.7607154 0 -2213.3374 0.27601803 -2213.0412 0.50460604 -2212.8271 1 -2213.3374
83 0.032624413 0.00016414467 0.0020866205 0.0031623348 0.01014434 0.5101467 0.51014204 1.7602604 0 -2213.3374 0.26052902 -2213.0671 0.50359805 -2212.8272 1 -2213.3374
93 0.011192577 1.2896354e-05 0.0014872247 0.0022320888 0.0057972393 0.5101135 0.51011119 1.7601271 0 -2213.3374 0.25446194 -2213.0775 0.50382575 -2212.8273 1 -2213.3374
96 0.0085623418 6.17654e-06 0.0013462454 0.0020157414 0.0050121139 0.51010903 0.51010716 1.7601216 0 -2213.3374 0.25374901 -2213.0787 0.50391708 -2212.8273 1 -2213.3374
and the corresponding log file for partition 0 looks like this:
LAMMPS (2 Aug 2023)
Processor partition = 0
OMP_NUM_THREADS environment is not set. Defaulting to 1 thread. (src/comm.cpp:98)
using 1 OpenMP thread(s) per MPI task
# NEB simulation of vacancy hopping in silicon crystal
units metal
atom_style atomic
atom_modify map array
boundary p p p
atom_modify sort 0 0.0
# coordination number cutoff
variable r equal 2.835
# diamond unit cell
variable a equal 5.431
lattice custom $a a1 1.0 0.0 0.0 a2 0.0 1.0 0.0 a3 0.0 0.0 1.0 basis 0.0 0.0 0.0 basis 0.0 0.5 0.5 basis 0.5 0.0 0.5 basis 0.5 0.5 0.0 basis 0.25 0.25 0.25 basis 0.25 0.75 0.75 basis 0.75 0.25 0.75 basis 0.75 0.75 0.25
lattice custom 5.431 a1 1.0 0.0 0.0 a2 0.0 1.0 0.0 a3 0.0 0.0 1.0 basis 0.0 0.0 0.0 basis 0.0 0.5 0.5 basis 0.5 0.0 0.5 basis 0.5 0.5 0.0 basis 0.25 0.25 0.25 basis 0.25 0.75 0.75 basis 0.75 0.25 0.75 basis 0.75 0.75 0.25
Lattice spacing in x,y,z = 5.431 5.431 5.431
region myreg block 0 4 0 4 0 4
#create_box 1 myreg
#create_atoms 1 region myreg
#mass 1 28.06
#write_data initial.sivac
read_data initial.sivac
Reading data file ...
orthogonal box = (0 0 0) to (21.724 21.724 21.724)
1 by 1 by 2 MPI processor grid
reading atoms ...
512 atoms
reading velocities ...
512 velocities
read_data CPU = 0.010 seconds
# make a vacancy
group Si type 1
512 atoms in group Si
group del id 300
1 atoms in group del
delete_atoms group del compress no
Deleted 1 atoms, new total = 511
group vacneigh id 174 175 301 304 306 331 337
7 atoms in group vacneigh
# choose potential
pair_style sw
pair_coeff * * Si.sw Si
Reading sw potential file Si.sw with DATE: 2007-06-11
# set up neb run
variable u uloop 20
# initial minimization to relax vacancy
displace_atoms all random 0.1 0.1 0.1 123456
Displacing atoms ...
minimize 1.0e-6 1.0e-4 1000 10000
Neighbor list info ...
update: every = 1 steps, delay = 0 steps, check = yes
max neighbors/atom: 2000, page size: 100000
master list distance cutoff = 5.77118
ghost atom cutoff = 5.77118
binsize = 2.88559, bins = 8 8 8
1 neighbor lists, perpetual/occasional/extra = 1 0 0
(1) pair sw, perpetual
attributes: full, newton on
pair build: full/bin/atomonly
stencil: full/bin/3d
bin: standard
Per MPI rank memory allocation (min/avg/max) = 4.113 | 4.114 | 4.114 Mbytes
Step Temp E_pair E_mol TotEng Press
0 0 -593.40319 0 -593.40319 355294.02
33 0 -2213.3343 0 -2213.3343 -3383.2606
Loop time of 0.0423295 on 2 procs for 33 steps with 511 atoms
97.9% CPU use with 2 MPI tasks x 1 OpenMP threads
Minimization stats:
Stopping criterion = energy tolerance
Energy initial, next-to-last, final =
-593.403188091472 -2213.33209897182 -2213.33426537417
Force two-norm initial, final = 1101.8254 0.16683659
Force max component initial, final = 334.49264 0.014961353
Final line search alpha, max atom move = 1 0.014961353
Iterations, force evaluations = 33 44
MPI task timing breakdown:
Section | min time | avg time | max time |%varavg| %total
---------------------------------------------------------------
Pair | 0.023435 | 0.030794 | 0.038153 | 4.2 | 72.75
Neigh | 0.00067653 | 0.00082372 | 0.00097091 | 0.0 | 1.95
Comm | 0.0021367 | 0.0096178 | 0.017099 | 7.6 | 22.72
Output | 0 | 0 | 0 | 0.0 | 0.00
Modify | 0 | 0 | 0 | 0.0 | 0.00
Other | | 0.001094 | | | 2.59
Nlocal: 255.5 ave 258 max 253 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Nghost: 1080.5 ave 1083 max 1078 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Neighs: 0 ave 0 max 0 min
Histogram: 2 0 0 0 0 0 0 0 0 0
FullNghs: 8689 ave 8777 max 8601 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Total # of neighbors = 17378
Ave neighs/atom = 34.007828
Neighbor list builds = 1
Dangerous builds = 0
reset_timestep 0
# only output atoms near vacancy
#dump events vacneigh custom 1000 dump.neb.sivac.$u id type x y z
fix 1 all neb 1.0
thermo 100
# run NEB for 2000 steps or to force tolerance
timestep 0.01
min_style quickmin
neb 0.0 0.01 100 100 10 final final.sivac verbosity default
Per MPI rank memory allocation (min/avg/max) = 2.989 | 2.989 | 2.989 Mbytes
Step Temp E_pair E_mol TotEng Press
0 0 -2213.3343 0 -2213.3343 -3383.2606
63 0.00010153187 -2213.3374 0 -2213.3374 -3383.3953
Loop time of 0.0424063 on 2 procs for 63 steps with 511 atoms
99.9% CPU use with 2 MPI tasks x 1 OpenMP threads
Minimization stats:
Stopping criterion = force tolerance
Energy initial, next-to-last, final =
-2213.33426537417 -2213.33736711116 -2213.3373695926
Force two-norm initial, final = 0.16683659 0.0047744861
Force max component initial, final = 0.014961353 0.00029110627
Final line search alpha, max atom move = 0 0
Iterations, force evaluations = 63 63
MPI task timing breakdown:
Section | min time | avg time | max time |%varavg| %total
---------------------------------------------------------------
Pair | 0.019724 | 0.024372 | 0.029021 | 3.0 | 57.47
Neigh | 0 | 0 | 0 | 0.0 | 0.00
Comm | 0.0029144 | 0.0075694 | 0.012224 | 5.4 | 17.85
Output | 0 | 0 | 0 | 0.0 | 0.00
Modify | 0.0073798 | 0.007426 | 0.0074723 | 0.1 | 17.51
Other | | 0.003038 | | | 7.17
Nlocal: 255.5 ave 260 max 251 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Nghost: 1080.5 ave 1085 max 1076 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Neighs: 0 ave 0 max 0 min
Histogram: 2 0 0 0 0 0 0 0 0 0
FullNghs: 8682 ave 8837 max 8527 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Total # of neighbors = 17364
Ave neighs/atom = 33.980431
Neighbor list builds = 0
Dangerous builds = 0
Per MPI rank memory allocation (min/avg/max) = 2.989 | 2.989 | 2.989 Mbytes
Step Temp E_pair E_mol TotEng Press
63 0 -2213.3374 0 -2213.3374 -3383.396
96 1.1818516e-06 -2213.3374 0 -2213.3374 -3383.3948
Loop time of 0.0229079 on 2 procs for 33 steps with 511 atoms
91.6% CPU use with 2 MPI tasks x 1 OpenMP threads
Minimization stats:
Stopping criterion = force tolerance
Energy initial, next-to-last, final =
-2213.3373695926 -2213.33738750829 -2213.337387544
Force two-norm initial, final = 0.0047744861 0.0013462454
Force max component initial, final = 0.00029110627 7.8360749e-05
Final line search alpha, max atom move = 0 0
Iterations, force evaluations = 33 33
MPI task timing breakdown:
Section | min time | avg time | max time |%varavg| %total
---------------------------------------------------------------
Pair | 0.010375 | 0.012454 | 0.014534 | 1.9 | 54.37
Neigh | 0 | 0 | 0 | 0.0 | 0.00
Comm | 0.0015189 | 0.0036039 | 0.0056889 | 3.5 | 15.73
Output | 0 | 0 | 0 | 0.0 | 0.00
Modify | 0.0049347 | 0.004958 | 0.0049814 | 0.0 | 21.64
Other | | 0.001891 | | | 8.26
Nlocal: 255.5 ave 260 max 251 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Nghost: 1080.5 ave 1085 max 1076 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Neighs: 0 ave 0 max 0 min
Histogram: 2 0 0 0 0 0 0 0 0 0
FullNghs: 8682 ave 8837 max 8527 min
Histogram: 1 0 0 0 0 0 0 0 0 1
Total # of neighbors = 17364
Ave neighs/atom = 33.980431
Neighbor list builds = 0
Dangerous builds = 0
Total wall time: 0:00:00
Now, from there since I found that adding the "verbosity" keyword to the "neb" command seems to fix one issue, I then add the "end" keyword to the "fix neb" command as:
fix 1 all neb 1.0 end last/efirst 1.0
thermo 100
# run NEB for 2000 steps or to force tolerance
timestep 0.01
min_style quickmin
neb 0.0 0.01 100 100 10 final final.sivac verbosity default
and when run, I get this error printed to the screen:
LAMMPS (2 Aug 2023)
Running on 4 partitions of processors
Reading NEB coordinate file(s) ...
Setting up regular NEB ...
[jcappola-Precision-Tower-7910:07399] *** An error occurred in MPI_Bcast
[jcappola-Precision-Tower-7910:07399] *** reported by process [4080271361,6]
[jcappola-Precision-Tower-7910:07399] *** on communicator MPI_COMM_WORLD
[jcappola-Precision-Tower-7910:07399] *** MPI_ERR_COMM: invalid communicator
[jcappola-Precision-Tower-7910:07399] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[jcappola-Precision-Tower-7910:07399] *** and potentially your MPI job)
[jcappola-Precision-Tower-7910:07389] 3 more processes have sent help message help-mpi-errors.txt / mpi_errors_are_fatal
[jcappola-Precision-Tower-7910:07389] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages
Again, all the screen.* files are blank and the log.lammps.* files are identical to the ones shown previously (where the previous error occured).
I've pulled down your latest bug fixes to the "fix_neb.cpp" and recompiled an otherwise clean version of LAMMPS (2 Aug 2023) - Update 2 including the MANYBODY and REPLICA package. It seems to have fixed the issue with the "end" keyword when run with multiple cores per replica, but the lack of a "verbosity" keyword in the "neb" command still gives the same MPI_Allgather error.
@jcappola thanks for the update. You can force unbuffered output (and thus avoid empty screen/log files) by adding the -nonbuf or short -nb command line flag. This was added specifically for debugging multi-replica jobs. 😉
I've identified an uninitialized data access in the neb.cpp code and refactored the flow of control a little bit to allow more code sharing (now that we require C++11 we can use a delegated constructor, which wasn't allowed when NEB was originally implemented).
The fix is added to the code. BTW, you do not have to manually apply the patch, but can just check out the "maintenance" branch (or download it as a snapshot from github) since these kinds of bugfixes are also backported and thus will show up in the third update to the stable version. Also see PR #4044
The verbosity bug does not show on my Linux box except when I explicitly test for uninitialized data with valgrind. By default allocated memory, and thus the variable triggering the output verbosity, are initialized to 0.
Good to know about -nb!
I've checked out the maintenance branch and compiled. Both of the bugs I reported are now fixed, thank you! This issue can now be closed.
@jcappola thanks for your feedback. While we do a lot of (automated) testing, there is nothing like bug reports from users to point out the oversights that are due to the many, many variants and way in how the different parts of LAMMPS can be used. While it is cool to have a software that can be used in many different ways like a set of Lego bricks and thus primarily limited by the creativity of its users, it also has the unique challenge that there are far too many permutations of combination of features and settings to test everything thoroughly. This is where people like you come in and which is what has made LAMMPS a much better and reliable software over the years. This has been particularly true since we moved the development process to GitHub and thus made it public. I hope cases like this keep you (and others that read this) motivated to keep reporting and suggesting.
The issue will be automatically closed once the pull request #4051 is merged into the develop branch.