CI enabled, reproducibility added, ready for release v1.2

.gitignore

@@ -9,3 +9,6 @@ build/LuaJIT-2.0.5*
 build/Makefile
 build/cmake_install.cmake
 build/openmm-7.3.0*
+build/CTestTestfile.cmake
+build/parRep
+build/Testing

.gitlab-ci.yml

+stages: 
+- build
+- test
+
+variables:
+    MAKE_COMMAND: "make -j 4"
+
+cache: # Directories that need to be kept between the build job and the test job
+    paths:
+    - build/
+
+#----------------------------------------------------------------
+# Linux build/test procedure
+#----------------------------------------------------------------
+
+# Build job
+parrep_build_ubuntu:
+    stage: build
+    tags:
+    - parrep-ubuntu-1604-amd64
+    script:
+    - cd build
+    - bash download_deps_and_run_cmake.sh
+    - make
+
+# Test job
+parrep_test_ubuntu:
+    stage: test
+    tags:
+    - parrep-ubuntu-1604-amd64
+    script:
+    - echo "$PWD"
+    - cd build
+    - ctest -V
+

CHANGELOG.md

+# v1.2.0
+
+Mar 08th, 2019
+
+The software was modified in order to allow reproducibility in some cases (the limiting factor is OpenMM which does not
+always provides deterministic output even when using the same seeds (!), see http://docs.openmm.org/latest/userguide/library.html#determinism ).
+
+The main executable now has 2 more command line options, '--inp-seeds [fname]' or '--out-seeds [fname]' for respectively loading seeds or writting seeds 
+from/to a unique binary file. See rand.hpp/rand.cpp, od the Doxygen doc for more details.
+
+These modifications now allow Continuous Integration (CI) on the infrastructure provided by INRIA : in mol/ci a small test case will be executed at each commit to the repository and compared to reference results.
+
+The two other minor modifications concern the Lua scripts:
+
+* "get_minimised_energy_crdvels" was added to the set of functions that the user can call from the Lua script : it simply combines in one call what 
+"get_minimised_energy" and "get_minimised_crdvels" already provided.
+
+* a extra optional parameter is available for the "simulation" parameters block when "simulation.algorithm" is "PARREP_FV" ; this parameter is
+"simulation.minAccumulatedObs" : it will enforce that at least minAccumulatedObs observations of an observable have already been accumulated
+before the convergence test is performed ; this may be useful if there is a risk of early pseudo-convergence for some of the observables when only a few samples have been accumulated.
+
+
+**Download sources:**
+
+https://gitlab.inria.fr/parallel-replica/gen.parRep/tags/v1.2.0
+
+or
+
+https://github.com/FHedin/gen.parRep/releases/tag/v1.2.0
+
 ----------------------------------------------
 
 # v1.1.0
 
+Nov 16th, 2018
+
 Finished the implementation of a "MD_interface" abstract class for future addition of different Molecular Dynamics engines.
 This required important changes to a significant part of the code.
 Optimizations to the MPI communications were also performed, therefore the minor version number
@@ -19,7 +51,9 @@ https://github.com/FHedin/gen.parRep/releases/tag/v1.1.0
 
 # v1.0.1
 
-simply added directory structure to readme file 
+Aug 16th, 2018
+
+Simply added directory structure to readme file.
 
 **Download sources:**
 
@@ -33,7 +67,9 @@ https://github.com/FHedin/gen.parRep/releases/tag/v1.0.1
 
 # v1.0.0
 
-Release of v1.0.0 : several bug fixes like memory leaks adressed, added input files for alanine
+Aug 13th, 2018
+
+Release of v1.0.0 : several bug fixes like memory leaks addressed, added input files for alanine
 dipeptide using transient propagation.
 
 **Download sources:**
@@ -52,7 +88,7 @@ July 6th, 2018
 
 First public release of the software.
 It is following the submission of the corresponding article on arXiv ; this is a release candidate 1.0.0-rc1,
-the 1.0.0 release should follow within the next weeks
+the 1.0.0 release should follow within the next weeks.
 
 **Download sources:**
 

CMakeLists.txt

@@ -15,6 +15,8 @@ if(NOT CMAKE_BUILD_TYPE)
       FORCE)
 endif()
 
+enable_testing()
+
 set(TGT "parRep" CACHE STRING "The executable name")
 
 project(${TGT} C CXX)
@@ -37,7 +39,7 @@ include
 include/dyna
 include/md
 include/utils
-# external headers (dependancies provided with this project)
+# external headers (dependencies provided with this project)
 external/sol2
 )
 
@@ -61,12 +63,10 @@ src/md/omm_interface.cpp
 # utilities files
 src/utils/mpi_utils.cpp
 src/utils/lua_interface.cpp
-# external dependancies provided with this project
+# external dependencies provided with this project
 external/luasqlite3/lsqlite3.c
 )
 
-add_executable(${TGT} ${SRCS})
-
 set(cWarnings "-Wall -Wextra")
 # more warnings for c++
 # set(cxxWarnings "-Wall -Wextra -Wformat=2 -Wshadow -Wconversion -Wuseless-cast")
@@ -90,4 +90,12 @@ set(CMAKE_CXX_FLAGS_RELEASE        "-O3    -march=native")
 set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O2 -g -march=native")
 set(CMAKE_CXX_FLAGS_MINSIZEREL     "-Os -s -march=native")
 
+add_executable(${TGT} ${SRCS})
+
 target_link_libraries(${TGT} ${OpenMM_LIBRARY} ${LUA_LIBRARIES} ${MPI_LIBRARIES} ${SQLITE3_LIBRARIES})
+
+add_test(
+NAME test_ci_ar7_gen.parRep
+COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/ci/run.ci.sh
+WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/ci
+)

LICENSE.txt

-Copyright (c) 2016-2018, Florent Hédin, Tony Lelièvre, and École des Ponts - ParisTech
+Copyright (c) 2016-2019, Florent Hédin, Tony Lelièvre, and École des Ponts - ParisTech,
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without modification,

README.md

-----------------------------------------------
-# gen.parRep (v1.1)
-----------------------------------------------
+# gen.parRep (v1.2)
 
 C++ implementation of the Generalized Parallel Replica algorithm
 
 Link to this repository : 
-  * https://gitlab.inria.fr/parallel-replica/gen.parRep
+
+* https://gitlab.inria.fr/parallel-replica/gen.parRep
 
 Link to the project gitlab page (includes more resources):
-  * https://gitlab.inria.fr/parallel-replica
+
+* https://gitlab.inria.fr/parallel-replica
 
 See the following articles:
-  * A generalized parallel replica dynamics: Binder, Lelièvre & Simpson, 2015: https://doi.org/10.1016/j.jcp.2015.01.002
-  * A new implementation of the Generalized Parallel Replica Dynamics targeting metastable biochemical systems: Hédin & Lelièvre, 2018: https://arxiv.org/abs/1807.02431
 
+* A generalized parallel replica dynamics: Binder, Lelièvre & Simpson, 2015: https://doi.org/10.1016/j.jcp.2015.01.002
+
+* gen.parRep: a first implementation of the Generalized Parallel Replica dynamics for the long time simulation of metastable biochemical systems: Hédin & Lelièvre, 2018: https://arxiv.org/abs/1807.02431 and https://doi.org/10.1016/j.cpc.2019.01.005
 
 Molecular dynamics is performed by using external codes linked to this program such as:
-  * OpenMM (https://github.com/pandegroup/openmm and https://simtk.org/home/openmm)
 
-----------------------------------------------
+* OpenMM (https://github.com/pandegroup/openmm and https://simtk.org/home/openmm)
+
+If you find any error, bug, or limitation, please open an issue on https://gitlab.inria.fr/parallel-replica/gen.parRep/issues so that the community can benefit from its correction ; also feel free to clone/fork the repo and perform the modifications yourself if you have the programming abilities to do so !
+
+Continuous Integration is performed on a test case system (see ./ci), ensuring that no release breaks the ability to compile the software; furthermore reference results from a previous run are available in ./ci/ref.sim in an attempt to achieve reproducibility whenever possible.
+
+[![pipeline status](https://gitlab.inria.fr/parallel-replica/gen.parRep/badges/test_ci/pipeline.svg)](https://gitlab.inria.fr/parallel-replica/gen.parRep/commits/master)
+
 ## DOCUMENTATION
-----------------------------------------------
 
 Please consult the Gitlab wiki for an overview of this software use: 
 
 https://gitlab.inria.fr/parallel-replica/gen.parRep/wikis/home
 
-Doxygen can be used for generating a code reference documentation within the docs directory:
-  * doxygen parRep.doxy 
+Doxygen can be used for generating a code reference documentation within the docs directory (particularly useful if you plan to modify/extend the software):
+
+* doxygen parRep.doxy 
 
 By default an HTML reference is generated (ready to read), and a docs/latex directory is also created, although the pdf version requires to be manually compiled (pdfLaTeX, see the Makefile in ./docs/latex).
 
-The Lua input files, in ./mol, used for starting computations, are self documented, and together
-with the bash submission files in the ./run directory should be enough for starting to use the program
+The Lua input files, in ./mol, used for starting simulations, are self documented, and together
+with the bash submission files in the ./run directory, it should be enough for starting to use the program.
+
+## DEPENDENCIES
+
+### Dependencies required before running the CMakeLists.txt script
 
-----------------------------------------------
-## DEPENDANCIES
-----------------------------------------------
+You will need an MPI development framework installed, compatible with the MPI 3.0 or newer standard : tested implementations :
 
-### Dependancies required before running the CMakeLists.txt script
+* Open MPI 1.10.2 (from Ubuntu 16.04 repositories)
 
-You will need an MPI development framework installed, compatible with the MPI 3.0 or newer standard : tested implementations : 
-  * Open MPI 1.10.2 (from Ubuntu 16.04 repositories)
 CMake will try to locate it automatically.
 
-You will need the SQLite3 dynamic library on your system: 
-See :
-  * https://sqlite.org/index.html
+You will need the SQLite3 dynamic library on your system. See :
 
-### Dependancies downloaded automatically via the script "build/download_dependencies.sh"
+* https://sqlite.org/index.html
 
-The script "build/download_dependencies.sh" can download and compile if required the OpenMM and LuaJIT dependancies.
+### Dependencies downloaded automatically via the script "build/download_deps_and_run_cmake.sh"
 
-You can however use your own version of the OpenMM library, or any Lua inplementation (either compiled yourself or downloaded somewhere).
+The script "build/download_deps_and_run_cmake.sh" will download and compile if required the OpenMM and LuaJIT Dependencies.
+
+You can however use your own version of the OpenMM library, or any Lua implementation (either compiled yourself or downloaded somewhere).
 
 #### Ressources (OpenMM): 
+
 * see https://simtk.org/home/openmm
+
 * and/or https://github.com/pandegroup/openmm
+
 * tested with version 7.0 to 7.3.
 
-You may need Nvidia CUDA or AMD OpenCL toolkit for enabling GPU acceleration ; see OpenMM documentation.
+You may need NVIDIA CUDA or AMD OpenCL toolkit for enabling GPU acceleration ; see OpenMM documentation.
 You may also need to edit CMakeLists.txt for specifying path to the include and lib directories of OpenMM.
 CMake will try to locate it automatically.
 
 #### Ressources (Lua/LuaJIT): 
 
 A Lua implementation compatible with Lua API version >= 5.1 is required :
-  * A release of Lua 5.x is usually already installed by default for recent linux versions, try to execute 'lua' and/or 'locate liblua'
-  * You can Download and compile the official implementation : http://www.lua.org/download.html
-  * The LuaJIT implementation can provide an important speedup : http://luajit.org/download.html
+
+* A release of Lua 5.x is usually already installed by default for recent linux versions, try to execute 'lua' and/or 'locate liblua'
+
+* You can Download and compile the official implementation : http://www.lua.org/download.html
+
+* The LuaJIT implementation can provide an important speedup : http://luajit.org/download.html
+
 Please download and compile any of the two. See below for hints for setting DCMAKE_PREFIX_PATH in case of a manual install.
 CMake will try to locate it automatically.
 
-### Dependancies provided in directory "external"
+### Dependencies provided in directory "external"
 
-The excellent Sol2 (header-only Lua<->c++ inteface) and LuaSQLite3 (a Lua/SQLite3 binding) dependancies are provided
+The excellent Sol2 (header-only Lua<->c++ interface) and LuaSQLite3 (a Lua/SQLite3 binding) Dependencies are provided
 in the './external' directory and are automatically included/compiled if required, and thus should not require extra
-configuration.
-See : 
-  * https://github.com/ThePhD/sol2
-  * http://lua.sqlite.org/index.cgi/home
+configuration. See :
+
+* https://github.com/ThePhD/sol2
+
+* http://lua.sqlite.org/index.cgi/home
 
-----------------------------------------------
 ## COMPILE & INSTALL
-----------------------------------------------
 
 C and C++ compiler compatible with the C99 and C++14 standards are required.
 Tested compilers:
-  * gcc/g++ 5.4.0 and 6.2.0 (from Ubuntu 16.04 repositories)
+
+* gcc/g++ 5.4.0 and 6.2.0 (from Ubuntu 16.04 repositories)
 
 Be sure to have CMake installed (http://www.cmake.org/), available on most repositories.
 Tested version :
-  * CMake 3.5.1 (from Ubuntu 16.04 repositories)
 
-Create a build directory and move to that directory: 
-  * mkdir build && cd build
+* CMake 3.5.1 (from Ubuntu 16.04 repositories)
+
+Create a build directory and move to that directory:
+
+* mkdir build && cd build
 
 For building a debug or release or an intermediate release with debug information, do:
-  * cmake -DCMAKE_BUILD_TYPE=Debug ..
-  * cmake -DCMAKE_BUILD_TYPE=RelWithDebInfo ..
-  * cmake -DCMAKE_BUILD_TYPE=Release ..         (default, with cpu targeting for the current compilation machine)
 
-Debug builds are possibly slower, usually more memory consuming, but useful when debugging with gdb or valgrind.
+* cmake -DCMAKE_BUILD_TYPE=Debug ..
+
+* cmake -DCMAKE_BUILD_TYPE=RelWithDebInfo ..
 
-If some dependancies were note detected (e.g. because they are not available within /usr/local) 
+* cmake -DCMAKE_BUILD_TYPE=Release ..         (default, with cpu targeting for the current compilation machine)
+
+Debug builds are possibly slower, usually more memory consuming, but useful when debugging with gdb or Valgrind.
+
+If some dependencies were note detected (e.g. because they are not available within /usr/local)
 it is required to add the path where the were installed to CMAKE_PREFIX_PATH, e.g.:
-  * cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="$HOME/bin/something" ..
+
+* cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="$HOME/bin/something" ..
 
 Then, once CMake built a Makefile without error, just execute the following in order to build the executable:
-  * make
 
-For a verbose make, use: 
-  * make VERBOSE=1
+* make
+
+For a verbose make, use:
+
+* make VERBOSE=1
 
 For a faster multi-core (on N cores) build, use:
-  * make -j N
 
-Please never edit the autogenerated Makefile, edit the CMakeLists.txt instead.
+* make -j N
+
+Please never edit the automatically generated Makefile, edit the CMakeLists.txt instead.
+
+For specifying another compiler on linux, for example clang, or a proprietary one like Intel icc (untested) :
+
+* CC=clang CXX=clang++ cmake ..
+
+* CC=icc CXX=icpc cmake ..
 
-For specifying another compiler on linux, for example clang, or a proprietary one like Intel icc (untested) : 
-  * CC=clang CXX=clang++ cmake ..
-  * CC=icc CXX=icpc cmake ..
- 
 In any case, you may need to edit the variables CMAKE_C_FLAGS_* and CMAKE_CXX_FLAGS_*
 for setting proper levels of optimization. By default 
 
-When using OpenMPI and forcing another C/C++ compilers version, the following might be required (possibly unsafe !): 
-  * OMPI_CC=gcc-5 OMPI_CXX=g++-5 CC=gcc-5 CXX=g++-5 ccmake ..
+When using OpenMPI and forcing another C/C++ compilers version, the following might be required (possibly unsafe !):
+
+* OMPI_CC=gcc-5 OMPI_CXX=g++-5 CC=gcc-5 CXX=g++-5 ccmake ..
 
-----------------------------------------------
 ## LICENSING (all files excepted subdirectory external and its content)
-----------------------------------------------
-Copyright (c) 2016-2018, Florent Hédin, Tony Lelièvre, and École des Ponts - ParisTech
+
+Copyright (c) 2016-2019, Florent Hédin, Tony Lelièvre, and École des Ponts - ParisTech
 All rights reserved.
 
 The 3-clause BSD license is applied to this software.
@@ -140,9 +165,7 @@ See LICENSE.txt
 
 See in ./external/* for licensing information concerning the embedded libraries provided in './external' 
 
-----------------------------------------------
 ## NOTES CONCERNING OpenMM
-----------------------------------------------
 
 The software should automatically detect the fastest OpenMM Platform available on your computer (i.e. CUDA, OpenCL, ...)
 If not it will run with the slow Reference platform : it is most probably because the OpenMM directory with the
@@ -151,18 +174,18 @@ If not it will run with the slow Reference platform : it is most probably becaus
 You may need to export the following OPENMM_PLUGIN_DIR environment variable to solve the problem : 
 
 For example for a custom installation in /home/$USER/bin/openmm
-  * export OPENMM_PLUGIN_DIR=$HOME/bin/openmm/lib/plugins
+
+* export OPENMM_PLUGIN_DIR=$HOME/bin/openmm/lib/plugins
 
 For setting the amount of cpu threads used by each MPI process use the following env. variable :
-  * export OPENMM_CPU_THREADS=1
 
-----------------------------------------------
+* export OPENMM_CPU_THREADS=1
+
 ## Directory structure
-----------------------------------------------
 
 **src** and **include** -> contains the C++ source and headers of the gen.parRep software.
 
-**build** -> directory where dependancies will be downloaded, and the program built.
+**build** -> directory where dependencies will be downloaded, and the program built.
 
 **external** -> contains source and headers of dependencies used by the gen.parRep software.
 
@@ -172,5 +195,6 @@ For setting the amount of cpu threads used by each MPI process use the following
 
 **mol** -> contains ready to use test systems, i.e. input files and OpenMM configurations of various molecular systems.
 
-**run** -> contains bash scripts demonstrating how to run the software using either mpirun or the SLURM scheduler; also contains three bash files for runing the software under the following debuggers/profilers: GDB, Valgrind and Scalasca.
+**run** -> contains bash scripts demonstrating how to run the software using either mpirun or the SLURM scheduler; also contains bash files for running the software under the following debuggers/profilers: GDB, Valgrind and Scalasca.
 
+**ci** -> contains a molecular system, an input script and reference data for performing continuous integration and non regression tests.

build/download_dependencies.sh → build/download_deps_and_run_cmake.sh

@@ -21,7 +21,7 @@ then
   fi
   tar xf LuaJIT-2.0.5.tar.gz
   cd LuaJIT-2.0.5
-  make PREFIX=$PWD/build -j 8 install
+  make PREFIX=$PWD/build install
   cd ..
 fi
 export LUAJIT_INSTALL_DIR=$PWD/LuaJIT-2.0.5/build

ci/.gitignore

+*.txt
+*.log
+*.db

ci/README.md

+# WARNING
+
+This is a simple test case executed when Continuous Integration tasks are performed ; it does not really make sense to use parRep on such a small test system !

ci/lj7/Integrator.xml

+<?xml version="1.0" ?>
+<Integrator constraintTolerance="1e-08" friction="10" randomSeed="123456789" stepSize=".001" temperature="15" type="LangevinIntegrator" version="1"/>

ci/lj7/State.xml

+<?xml version="1.0" ?>
+<State openmmVersion="7.3" time="999.9999999832651" type="State" version="1">
+	<PeriodicBoxVectors>
+		<A x="20" y="0" z="0"/>
+		<B x="0" y="20" z="0"/>
+		<C x="0" y="0" z="20"/>
+	</PeriodicBoxVectors>
+	<Positions>
+		<Position x="9.983385059299602" y="10.174980427137598" z="10.020986348782928"/>
+		<Position x="9.770224100810191" y="10.037570850276143" z="10.293565548378014"/>
+		<Position x="9.59030488821879" y="10.02885267592082" z="9.931518707870099"/>
+		<Position x="9.648648572516473" y="10.349048230249956" z="10.107352368306797"/>
+		<Position x="9.766183439034616" y="10.306491926935744" z="9.758201173961519"/>
+		<Position x="9.93467672489858" y="9.95859189806722" z="9.682105588219425"/>
+		<Position x="9.906861537781493" y="9.817471151536196" z="10.038531569331555"/>
+	</Positions>
+	<Velocities>
+		<Velocity x=".06812021378266309" y="-.014536461211633878" z="-.05354852211603145"/>
+		<Velocity x="-.0003575163347591115" y=".004253822229571824" z=".10209696385565792"/>
+		<Velocity x=".01955586418134203" y=".01921201910093373" z="-.003143989026455074"/>
+		<Velocity x=".06026768071265565" y=".04230138984162579" z=".07773255456200445"/>
+		<Velocity x="-.05115033391156487" y=".045557891198200196" z=".03101618878886825"/>
+		<Velocity x="-.028826659962177814" y=".059965265059247486" z="-.07743277297400653"/>
+		<Velocity x=".06604694795520061" y="-.02419439117318234" z=".015470123640781708"/>
+	</Velocities>
+</State>

ci/lj7/System.xml

+<?xml version="1.0" ?>
+<System openmmVersion="7.3" type="System" version="1">
+	<PeriodicBoxVectors>
+		<A x="20" y="0" z="0"/>
+		<B x="0" y="20" z="0"/>
+		<C x="0" y="0" z="20"/>
+	</PeriodicBoxVectors>
+	<Particles>
+		<Particle mass="39.948"/>
+		<Particle mass="39.948"/>
+		<Particle mass="39.948"/>
+		<Particle mass="39.948"/>
+		<Particle mass="39.948"/>
+		<Particle mass="39.948"/>
+		<Particle mass="39.948"/>
+	</Particles>
+	<Constraints/>
+	<Forces>
+		<Force alpha="0" cutoff="1" dispersionCorrection="0" ewaldTolerance=".0005" forceGroup="0" ljAlpha="0" ljnx="0" ljny="0" ljnz="0" method="0" nx="0" ny="0" nz="0" recipForceGroup="-1" rfDielectric="78.3" switchingDistance="-1" type="NonbondedForce" useSwitchingFunction="0" version="3">
+			<GlobalParameters/>
+			<ParticleOffsets/>
+			<ExceptionOffsets/>
+			<Particles>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+				<Particle eps=".99607255958" q="0" sig=".3405"/>
+			</Particles>
+			<Exceptions/>
+		</Force>
+		<Force forceGroup="0" frequency="100" type="CMMotionRemover" version="1"/>
+	</Forces>
+</System>

ci/lj7/omm_LJ7.py

+from simtk.openmm.app import *
+from simtk.openmm import *
+from simtk.unit import *
+
+import numpy as np
+from math import *
+from sys import stdout
+
+platform = Platform.getPlatformByName('Reference')
+
+# periodic conditions
+# periodic vectors
+a = Vec3(20.0,0.0,0.0)
+b = Vec3(0.0,20.0,0.0)
+c = Vec3(0.0,0.0,20.0)
+
+center = Vec3(10.,10.,10.)
+
+print("a = ",a)
+print("b = ",b)
+print("c = ",c)
+print("center = ",center)
+
+natoms = 7
+mass = 39.948*amu
+temperature = 15.0*kelvin
+
+integrator = LangevinIntegrator(temperature, 10/picosecond, 0.001*picosecond)
+integrator.setConstraintTolerance(1e-8)
+
+integrator.setRandomNumberSeed(123456789)
+
+nb = NonbondedForce()
+nb.setNonbondedMethod(NonbondedForce.NoCutoff)
+#nb.setUseSwitchingFunction(True)
+#nb.setSwitchingDistance(1.0*nanometer)
+#nb.setCutoffDistance(1.2*nanometer)
+nb.setUseDispersionCorrection(False)
+
+# see http://www.sklogwiki.org/SklogWiki/index.php/Argon for argon LJ parameters
+k_B = 0.0083144621*(kilojoule/(mole*kelvin))
+lj_epsilon = 119.8*kelvin * k_B # in kJ/mol
+lj_sigma = 0.3405*nanometer  # in nm
+
+# All seven particles will initially be placed on a 2D plane
+# Each particle is by construction at distance r_min of another atom
+# r_min = lj_sigma * 2^(1./6.) is the distance at which the LJ potential is minimum
+# See on the right for the numbering of particles
+#
+#     *   *               3   2
+#   *   *   *           4   0   1
+#     *   *               5   6
+#
+r_min = 2.**(1./6.) * lj_sigma._value
+xpos  = r_min*cos(pi/3.)
+ypos  = r_min*sin(pi/3.)
+
+iniPos = np.zeros((natoms,3))
+iniPos[0,:] = (center[0],center[1],center[2])
+iniPos[1,:] = (center[0]+r_min,center[1],center[2])
+iniPos[2,:] = (center[0]+xpos,center[1]+ypos,center[2])
+iniPos[3,:] = (center[0]-xpos,center[1]+ypos,center[2])
+iniPos[4,:] = (center[0]-r_min,center[1],center[2])
+iniPos[5,:] = (center[0]-xpos,center[1]-ypos,center[2])
+iniPos[6,:] = (center[0]+xpos,center[1]-ypos,center[2])
+
+system = System()
+for i in range(natoms):
+  system.addParticle(mass)
+  nb.addParticle(0.0,lj_sigma,lj_epsilon)
+  
+system.setDefaultPeriodicBoxVectors(a,b,c)
+system.addForce(nb)
+system.addForce(CMMotionRemover(100))
+
+# generate basic topology object to use with simulation class (and to be able to write pdb files)
+top = Topology()
+top.setPeriodicBoxVectors([a,b,c])
+chain = top.addChain("A")
+for i in range(natoms):
+  top.addResidue("AR",chain)
+for r in top.residues():
+  top.addAtom("AR",Element.getBySymbol("Ar"),r)
+
+
+simulation = Simulation(top,system,integrator,platform)#,platformProperties)
+
+simulation.context.setPositions(iniPos)
+
+state = simulation.context.getState(getPositions=True,getVelocities=True,enforcePeriodicBox=True)
+with open("start.pdb", 'w') as f:
+  PDBFile.writeFile(top,state.getPositions(),f)
+  
+print('Minimizing energy...')
+simulation.minimizeEnergy(1e-5)
+
+simulation.context.setVelocitiesToTemperature(temperature)
+
+state = simulation.context.getState(getPositions=True,getVelocities=True,enforcePeriodicBox=True)
+with open("minim.pdb", 'w') as f:
+  PDBFile.writeFile(top,state.getPositions(),f)
+  
+nsteps   = 1e6 
+# save each 1 ps
+saveFreq = 1000
+
+simulation.reporters.append(StateDataReporter(stdout, saveFreq, step=True, 
+    time=True, potentialEnergy=True, kineticEnergy=True, totalEnergy=True,
+    temperature=True, progress=True,
+    remainingTime=True, speed=True,
+    totalSteps=nsteps, separator='\t'))
+
+simulation.reporters.append(DCDReporter(file='sim.dcd', reportInterval=saveFreq, enforcePeriodicBox=True))
+
+print("Running dynamics...")
+simulation.step(nsteps)
+
+state = simulation.context.getState(getPositions=True,getVelocities=True,enforcePeriodicBox=True)
+with open("1ns.pdb", 'w') as f:
+  PDBFile.writeFile(top,state.getPositions(),f)
+  
+#serialize the state to xml for load in parRep later
+print('Serializing state : it contains coordinates, velocities, etc. for parRep...')
+f = open('State.xml','w')
+f.write(XmlSerializer.serialize(state))
+f.close()
+
+#serialize the system to xml for load in parRep later
+print('Serializing system : it contains all definition of forces for parRep...')
+f = open('System.xml','w')
+f.write(XmlSerializer.serialize(system))
+f.close()
+
+#serialize the integrator to xml for load in parRep later
+print('Serializing integrator for parRep...')
+f = open('Integrator.xml','w')
+f.write(XmlSerializer.serialize(integrator))
+f.close()
+

ci/ref.sim/README.md

+# Software used for producing this output : 
+
+sqlite3 : 3.22.0 2018-01-22 18:45:57 0c55d179733b46d8d0ba4d88e01a25e10677046ee3da1d5b1581e86726f2alt1
+
+gcc/g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0
+
+Luajit : LuaJIT 2.0.5, download and compiled from script download_dependcies in the build directory.
+
+OpenMM : openmm-7.3.0-py27_cuda92_rc_1.tar.bz2, bin version downloaded from script download_dependcies in the build directory.
+
+MPI : mpirun (Open MPI) 2.1.1, from repository
+
+# Linux (Ubuntu 18.04): 
+
+$ lsb_release -a
+LSB Version:    core-9.20170808ubuntu1-noarch:printing-9.20170808ubuntu1-noarch:security-9.20170808ubuntu1-noarch
+Distributor ID: Ubuntu
+Description:    Ubuntu 18.04.2 LTS
+Release:        18.04
+Codename:       bionic
+
+$ uname -a
+Linux pl267-pro 4.15.0-45-generic #48-Ubuntu SMP Tue Jan 29 16:28:13 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

ci/run.ci.sh

+#!/bin/bash
+
+rm -f *.db *.log *.txt
+
+export OPENMM_CPU_THREADS=1
+export OPENMM_PLUGIN_DIR=$PWD/../build/openmm-7.3.0/lib/plugins
+
+mpirun -np 8 $PWD/../build/parRep -i input_generalized_parRep.lua -log warn -o out.txt -e err.txt --inp-seeds ./ref.sim/seeds.bin
+
+MPI_RET_CODE=$?
+
+# compare the db file from the run and the reference, they should be the same
+sqldiff lj7.db ref.sim/lj7.db > diff.txt
+
+SQL_RET_CODE=$?
+
+if [ -s "diff.txt" ]
+then 
+  TEST_RET_CODE=-1
+else
+  TEST_RET_CODE=0
+fi
+
+cat out.*.txt
+
+# success only if there was no error code from any of the previous commands
+exit $(($MPI_RET_CODE|$SQL_RET_CODE|$TEST_RET_CODE))

include/dyna/GelmanRubin.hpp

@@ -4,7 +4,7 @@
  * \author Florent Hédin
  * \author Tony Lelièvre
  * \author École des Ponts - ParisTech
- * \date 2016-2018
+ * \date 2016-2019
  */
 
 #ifndef GELMANRUBIN_HPP_INCLUDED
@@ -34,13 +34,11 @@ public:
 
   /**
   * @brief Constructor for a Gelman Rubin Analysis object; requires at least the number of chains (replica) as first argument.
-  *        The second argument allows to discard the first n observations when averaging, default is zero and thus disables 
-  *        discarding.  
   * 
   * @param _num_chains The number of G-R chains (same as the number of replicas usually).
-  * @param _discard Discards the first n observations, as usually there won't be convergence at the beginning.
+  * @param _no_convergence_if_less_than Optional argument : if set there won't be convergence check until at least _no_convergence_if_less_than observations have been accumulated for each observable : this may hel avoiding pseudo-convergence.
   */
-  GelmanRubinAnalysis(const uint32_t _num_chains, const uint32_t _discard=0);
+  GelmanRubinAnalysis(const uint32_t _num_chains, const uint32_t _no_convergence_if_less_than=0);
   
   
   /**
@@ -195,8 +193,9 @@ private:
   std::map<std::string,double> tolerance;             ///< Allows one tolerance treshold per observable type
   std::map<std::string,std::vector<double>> ratio;    ///< Allows one ratio vector per observable type
   
-  uint32_t num_chains;                      ///< The total number of chains
-  uint32_t discard_first;                   ///< A given number of records to ignore at the beginning
+  uint32_t num_chains = 0;                          ///< The total number of chains
+  uint32_t min_num_observations_before_check = 0;   ///< There won't be convergence check until at least this amount ob observations have been accumulated for each obseravble
+  
 };
 
 

include/dyna/observable.hpp

@@ -4,7 +4,7 @@
  * \author Florent Hédin
  * \author Tony Lelièvre
  * \author École des Ponts - ParisTech
- * \date 2016-2018
+ * \date 2016-2019
  */
 
 #ifndef OBSERVABLE_HPP_INCLUDED