scnc
/
qmcpack
mirror of https://github.com/QMCPACK/qmcpack.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

finally a decent code block 

git-svn-id: https://subversion.assembla.com/svn/qmcdev/trunk@4869 e5b18d87-469d-4833-9cc0-8cdfa06e9491
This commit is contained in:
David/ChangMo Yang 2010-06-08 05:58:05 +00:00
parent b2537f9580
commit 2adba57969
7 changed files with 378 additions and 420 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
docs/usrdoc/3dheg.s000.xml Normal file
View File

@ -0,0 +1,31 @@
<qmcsystem>
<simulationcell>
<!-- Note 1.
change condition for different number of electrons. Only closed-shell
systems are supported for the moment. The allowed numbers are:
14 38 54 66 114 162 ....
-->
<parameter name="rs" condition="14">5</parameter>
<parameter name="bconds">p p p</parameter>
<parameter name="LR\_dim\_cutoff">6</parameter>
</simulationcell>
<particleset name="e" random="yes">
<group name="u" size="7">
<parameter name="charge">-1</parameter>
</group>
<group name="d" size="7">
<parameter name="charge">-1</parameter>
</group>
</particleset>
<hamiltonian name="h0" type="generic" target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
</hamiltonian>
<wavefunction name="psi0" target="e">
<!-- Note 2.
change shell according to the parameter/@condition
14(1) 38(2) 54 (3) 66 (4) 114 (5) 162 (6) ...
-->
<determinantset type="electron-gas" shell="1" randomize="true"/>
<jastrow name="Jee" type="Two-Body" function="rpa"/>
</wavefunction>
</qmcsystem>

62
docs/usrdoc/cBNprim.in Normal file
View File

@ -0,0 +1,62 @@
# Cubic Boron Nitride : primitive cell orbitals
#
# Definition of the unit cell
# BN lattice constant near experimental volume
acell 3*6.839904
# FCC lattice vectors
rprim 0.5 0.5 0.0
0.0 0.5 0.5
0.5 0.0 0.5
# Definition of the atom types
ntypat 2 # There is only one type of atom
znucl 5 7 # The keyword "znucl" refers to the atomic number of the
# possible type(s) of atom. The pseudopotential(s)
# mentioned in the "files" file must correspond
# to the type(s) of atom. Here, the only type is Silicon.
# Definition of the atoms
natom 2 # There are two atoms
# They both of type boron and nitrogen
typat 1 2
xred
0.0000 0.0000 0.0000 # B1
0.2500 0.2500 0.2500 # N1
# Definition of the planewave basis set
ecut 200.0 # Maximal kinetic energy cut-off, in Hartree
# Allow non-primitive unit cell
chkprim 0
#Definition of the k-point grid
##################################
# USE THESE FOR CASINO OUTPUT!!! #
##################################
kptopt 0 # Manual k-points
nkpt 1
istwfk 1
wtk 1
kpt
0.000 0.000 0.000
prtwf 1
# Definition of the SCF procedure
nstep 100 # Maximal number of SCF cycles
toldfe 1.0d-12 # Will stop when, twice in a row, the difference
# between two consecutive evaluations of total energy
# differ by less than toldfe (in Hartree)
diemac 2.0 # Although this is not mandatory, it is worth to
# precondition the SCF cycle. The model dielectric
# function used as the standard preconditioner
# is described in the "dielng" input variable section.
# Here, we follow the prescription for bulk silicon.
# ixc 23 # Wu-Cohen
ixc 7 # LDA

142
docs/usrdoc/cBNprim.qmc.xml Normal file
View File

@ -0,0 +1,142 @@
<?xml version="1.0"?>
<simulation>
<project id="BN2\_prim" series="0">
<application name="qmcapp" role="molecu" class="serial" version="0.2">
</application>
</project>
<random seed="13"/>
<qmcsystem dim="3">
<simulationcell>
<parameter name="scale"> 6.839904 </parameter>
<parameter name="lattice">
0.5 0.5 0.0
0.0 0.5 0.5
0.5 0.0 0.5
</parameter>
<parameter name="bconds">p p p</parameter>
<parameter name="LR\_dim\_cutoff">20.0</parameter>
</simulationcell>
<particleset name="e" random="yes">
<group name="u" size="4">
<parameter name="charge">-1</parameter>
</group>
<group name="d" size="4">
<parameter name="charge">-1</parameter>
</group>
</particleset>
<particleset name="i" size="2">
<group name="B">
<parameter name="charge">3</parameter>
<parameter name="valence">3</parameter>
<parameter name="atomicnumber">5</parameter>
</group>
<group name="N">
<parameter name="charge">5</parameter>
<parameter name="valence">5</parameter>
<parameter name="atomicnumber">7</parameter>
</group>
<attrib name="position" datatype="posArray" condition="1">
0.000 0.000 0.000
0.250 0.250 0.250
</attrib>
<attrib name="ionid" datatype="stringArray">
B N
</attrib>
</particleset>
<wavefunction name="psi0" target="e">
<jastrow name="J1" type="One-Body" function="Bspline" print="yes" source="i">
<correlation elementType="B" cusp="0.0" size="2">
<coefficients id="b" type="Array">
0 0
</coefficients>
</correlation>
<correlation elementType="N" cusp="0.0" size="6">
<coefficients id="n" type="Array">
0 0
</coefficients>
</correlation>
</jastrow>
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
<correlation pairType="uu" speciesA="u" speciesB="u" size="2">
<coefficients id="uu" type="Array">
0 0
</coefficients>
</correlation>
<correlation pairType="ud" speciesA="u" speciesB="d" size="2">
<coefficients id="ud" type="Array">
0 0
</coefficients>
</correlation>
</jastrow>
<determinantset type="einspline" href="cBNprim.h5" sort="1">
<basisset/>
<slaterdeterminant>
<determinant id="updet" size="4">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
<determinant id="downdet" size="4">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
</slaterdeterminant>
</determinantset>
</wavefunction>
<hamiltonian name="h0" type="generic" target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
<pairpot type="pseudo" name="PPot" source="i" wavefunction="psi0"
format="xml">
<pseudo elementType="B" href="B.BFD.fsatom.xml" format="xml"/>
<pseudo elementType="N" href="N.BFD.fsatom.xml" format="xml"/>
</pairpot>
<constant name="IonIon" type="coulomb" source="i" target="i"/>
</hamiltonian>
</qmcsystem>
<!-- Optimize -->
<loop max="2">
<qmc method="optimize" move="pbyp">
<parameter name="blocks"> 1000 </parameter>
<parameter name="steps"> 10 </parameter>
<parameter name="timestep"> 0.7 </parameter>
<parameter name="walkers"> 8 </parameter>
<parameter name="samples"> 1000 </parameter>
<cost name="energy"> 0.0 </cost>
<cost name="variance"> 1.0 </cost>
<cost name="difference"> 0.0 </cost>
<parameter name="minWalkers"> 0.4 </parameter>
<parameter name="useWeight"> no </parameter>
<optimizer method="cg">
<parameter name="max\_steps"> 15 </parameter>
<parameter name="tolerance"> 1e-6 </parameter>
<parameter name="stepsize"> 0.02 </parameter>
<parameter name="friction"> 5 </parameter>
<parameter name="epsilon"> 0.0 </parameter>
</optimizer>
<parameter name="power"> 2 </parameter>
<parameter name="correlation"> 0.01 </parameter>
</qmc>
</loop>
<!-- WARMUP -->
<qmc method="vmc" move="pbyp">
<estimator name="LocalEnergy"/>
<parameter name="walkers"> 8 </parameter>
<parameter name="warmupSteps"> 100 </parameter>
<parameter name="blocks"> 160 </parameter>
<parameter name="steps"> 10 </parameter>
<parameter name="samples"> 1280 </parameter>
<parameter name="timestep"> 0.5 </parameter>
</qmc>
<qmc method="dmc" move="pbyp">
<parameter name="blocks"> 20000 </parameter>
<parameter name="steps"> 10 </parameter>
<parameter name="timestep"> 0.01 </parameter>
<parameter name="nonlocalmoves"> yes </parameter>
<parameter name="warmupSteps"> 500 </parameter>
</qmc>
</simulation>

295
docs/usrdoc/qmcpack-u-ex.tex
View File

@ -23,8 +23,8 @@ After downloading these tools, do \icode{chmod u+x wfconv ppconvert} to mark the
\subsection*{Convert the pseudopotentials}
[check this, perhaps also explaining what each line in the file means]
Copy the following pseudopotential in GAMESS format into a file "B.BFD.gamess":
\begin{lstcpp}
Copy the following pseudopotential in GAMESS format into a file \iconsole{B.BFD.gamess}:
\begin{code}
B-QMC GEN 2 1
3
3.00000000 1 5.40423964
@ -32,10 +32,10 @@ B-QMC GEN 2 1
-11.86640633 2 4.48974455
1
15.49737620 2 3.43781634
\end{lstcpp}
\end{code}
Likewise, the following into "N.BFD.gamess":
\begin{verbatim}
Likewise, the following into \iconsole{N.BFD.gamess}:
\begin{code}
N-QMC GEN 2 1
3
5.00000000 1 9.23501007
@ -43,15 +43,15 @@ N-QMC GEN 2 1
-30.18893534 2 7.34486070
1
31.69720409 2 6.99536540
\end{verbatim}
\end{code}
We will now convert the pseudopotentials into the FHI format used by ABINIT and the FSatom XML format used by QMCPACK. Put ppconvert into a directory in your PATH. Then execute
\begin{verbatim}
\begin{code}
ppconvert --gamess_pot B.BFD.gamess --s_ref "1s(2)2p(1)" --p_ref "1s(2)2p(1)" \
--fhi B.BFD.fhi --xml B.BFD.fsatom.xml
ppconvert --gamess_pot N.BFD.gamess --s_ref "1s(2)2p(3)" --p_ref "1s(2)2p(3)" \
--fhi N.BFD.fhi --xml N.BFD.fsatom.xml
\end{verbatim}
\end{code}
\begin{itemize}
\item{} The first argument given is the input pseudopotential file.
\item{} The second and third arguments give the reference state for forming Kleinmann-Bylander projectors.
@ -64,74 +64,11 @@ ppconvert --gamess_pot N.BFD.gamess --s_ref "1s(2)2p(3)" --p_ref "1s(2)2p(3)" \
\subsection*{Generating orbitals with ABINIT}
ABINIT (http://www.abinit.org) is a general-purpose plane-wave DFT code which supports pseudopotential and PAW calculations. It is well-documented, full-featured, and has a vibrant community support forum.
We will begin with a primitive cell of c-BN. Copy the following into cBNprim.in:
\begin{verbatim}
# Cubic Boron Nitride : primitive cell orbitals
#
# Definition of the unit cell
# BN lattice constant near experimental volume
acell 3*6.839904
We will begin with a primitive cell of c-BN. Copy the following into \iconsole{cBNprim.in}:
\codesrc{cBNprim.in}
# FCC lattice vectors
rprim 0.5 0.5 0.0
0.0 0.5 0.5
0.5 0.0 0.5
# Definition of the atom types
ntypat 2 # There is only one type of atom
znucl 5 7 # The keyword "znucl" refers to the atomic number of the
# possible type(s) of atom. The pseudopotential(s)
# mentioned in the "files" file must correspond
# to the type(s) of atom. Here, the only type is Silicon.
# Definition of the atoms
natom 2 # There are two atoms
# They both of type boron and nitrogen
typat 1 2
xred
0.0000 0.0000 0.0000 # B1
0.2500 0.2500 0.2500 # N1
# Definition of the planewave basis set
ecut 200.0 # Maximal kinetic energy cut-off, in Hartree
# Allow non-primitive unit cell
chkprim 0
#Definition of the k-point grid
##################################
# USE THESE FOR CASINO OUTPUT!!! #
##################################
kptopt 0 # Manual k-points
nkpt 1
istwfk 1
wtk 1
kpt
0.000 0.000 0.000
prtwf 1
# Definition of the SCF procedure
nstep 100 # Maximal number of SCF cycles
toldfe 1.0d-12 # Will stop when, twice in a row, the difference
# between two consecutive evaluations of total energy
# differ by less than toldfe (in Hartree)
diemac 2.0 # Although this is not mandatory, it is worth to
# precondition the SCF cycle. The model dielectric
# function used as the standard preconditioner
# is described in the "dielng" input variable section.
# Here, we follow the prescription for bulk silicon.
# ixc 23 # Wu-Cohen
ixc 7 # LDA
\end{verbatim}
Copy the following into cBNprim.files:
\begin{verbatim}
Copy the following into \iconsole{cBNprim.files}:
\begin{code}
cBNprim.in
cBNprim.out
cBNprim.xi
@ -139,181 +76,35 @@ cBNprim.xo
cBNprim_
B.BFD.fhi
N.BFD.fhi
\end{verbatim}
\end{code}
Now, run
\begin{verbatim}
\begin{console}
abinis < cBNprim.files
\end{verbatim}
\end{console}
\subsection*{Converting the orbitals}
With wfconv in your PATH, run
\begin{verbatim}
\begin{console}
wfconv --eshdf cBNprim.h5 cBNprim.xo_WFK
\end{verbatim}
\end{console}
This will generated an orbital file in the ESHDF format that QMCPACK reads.
N.B. The GPU version of QMCPACK uses an older format for the orbital file. To generate orbitals for the GPU code, do instead
\begin{verbatim}
\begin{console}
wfconv --spline --qmcPACK cBNprim.h5 cBNprim.xo_WFK
\end{verbatim}
\end{console}
The newer CPU code also can read this format, but we are trying to deprecate it. We intend to remerge the GPU and CPU versions in the near future.
\subsection*{Running QMCPACK}
Copy the following to ``cBNprim.qmc.xml'':
\begin{lstcpp}
<?xml version="1.0"?>
<simulation>
<project id="BN2_prim" series="0">
<application name="qmcapp" role="molecu" class="serial" version="0.2">
</application>
</project>
<random seed="13"/>
<qmcsystem dim="3">
<simulationcell>
<parameter name="scale"> 6.839904 </parameter>
<parameter name="lattice">
0.5 0.5 0.0
0.0 0.5 0.5
0.5 0.0 0.5
</parameter>
<parameter name="bconds">p p p</parameter>
<parameter name="LR_dim_cutoff">20.0</parameter>
</simulationcell>
<particleset name="e" random="yes">
<group name="u" size="4">
<parameter name="charge">-1</parameter>
</group>
<group name="d" size="4">
<parameter name="charge">-1</parameter>
</group>
</particleset>
<particleset name="i" size="2">
<group name="B">
<parameter name="charge">3</parameter>
<parameter name="valence">3</parameter>
<parameter name="atomicnumber">5</parameter>
</group>
<group name="N">
<parameter name="charge">5</parameter>
<parameter name="valence">5</parameter>
<parameter name="atomicnumber">7</parameter>
</group>
<attrib name="position" datatype="posArray" condition="1">
0.000 0.000 0.000
0.250 0.250 0.250
</attrib>
<attrib name="ionid" datatype="stringArray">
B N
</attrib>
</particleset>
<wavefunction name="psi0" target="e">
<jastrow name="J1" type="One-Body" function="Bspline" print="yes" source="i">
<correlation elementType="B" cusp="0.0" size="2">
<coefficients id="b" type="Array">
0 0
</coefficients>
</correlation>
<correlation elementType="N" cusp="0.0" size="6">
<coefficients id="n" type="Array">
0 0
</coefficients>
</correlation>
</jastrow>
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
<correlation pairType="uu" speciesA="u" speciesB="u" size="2">
<coefficients id="uu" type="Array">
0 0
</coefficients>
</correlation>
<correlation pairType="ud" speciesA="u" speciesB="d" size="2">
<coefficients id="ud" type="Array">
0 0
</coefficients>
</correlation>
</jastrow>
<determinantset type="einspline" href="cBNprim.h5" sort="1">
<basisset/>
<slaterdeterminant>
<determinant id="updet" size="4">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
<determinant id="downdet" size="4">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
</slaterdeterminant>
</determinantset>
</wavefunction>
<hamiltonian name="h0" type="generic" target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
<pairpot type="pseudo" name="PPot" source="i" wavefunction="psi0"
format="xml">
<pseudo elementType="B" href="B.BFD.fsatom.xml" format="xml"/>
<pseudo elementType="N" href="N.BFD.fsatom.xml" format="xml"/>
</pairpot>
<constant name="IonIon" type="coulomb" source="i" target="i"/>
</hamiltonian>
</qmcsystem>
<!-- Optimize -->
<loop max="2">
<qmc method="optimize" move="pbyp">
<parameter name="blocks"> 1000 </parameter>
<parameter name="steps"> 10 </parameter>
<parameter name="timestep"> 0.7 </parameter>
<parameter name="walkers"> 8 </parameter>
<parameter name="samples"> 1000 </parameter>
<cost name="energy"> 0.0 </cost>
<cost name="variance"> 1.0 </cost>
<cost name="difference"> 0.0 </cost>
<parameter name="minWalkers"> 0.4 </parameter>
<parameter name="useWeight"> no </parameter>
<optimizer method="cg">
<parameter name="max_steps"> 15 </parameter>
<parameter name="tolerance"> 1e-6 </parameter>
<parameter name="stepsize"> 0.02 </parameter>
<parameter name="friction"> 5 </parameter>
<parameter name="epsilon"> 0.0 </parameter>
</optimizer>
<parameter name="power"> 2 </parameter>
<parameter name="correlation"> 0.01 </parameter>
</qmc>
</loop>
<!-- WARMUP -->
<qmc method="vmc" move="pbyp">
<estimator name="LocalEnergy"/>
<parameter name="walkers"> 8 </parameter>
<parameter name="warmupSteps"> 100 </parameter>
<parameter name="blocks"> 160 </parameter>
<parameter name="steps"> 10 </parameter>
<parameter name="samples"> 1280 </parameter>
<parameter name="timestep"> 0.5 </parameter>
</qmc>
Copy the following to \iconsole{cBNprim.qmc.xml}:
\codesrc{cBNprim.qmc.xml}
<qmc method="dmc" move="pbyp">
<parameter name="blocks"> 20000 </parameter>
<parameter name="steps"> 10 </parameter>
<parameter name="timestep"> 0.01 </parameter>
<parameter name="nonlocalmoves"> yes </parameter>
<parameter name="warmupSteps"> 500 </parameter>
</qmc>
</simulation>
\end{lstcpp}
More from
\begin{verbatim}
http://cms.mcc.uiuc.edu/qmcpack/index.php/Bulk_solid_calculations_with_DFT_orbitals
http://qmcpack.cmscc.org/tutorials/bulk-solid
\end{verbatim}
More from\\
\url{http://cms.mcc.uiuc.edu/qmcpack/index.php/Bulk_solid_calculations_with_DFT_orbitals}\\
\url{http://qmcpack.cmscc.org/tutorials/bulk-solid}
\section{Liquid helium}
\section{3D homogeneous electron gas}
Reviews on homogeneous electron gas can be found in standard solid-state physics or electronic structure textbooks. A particularly relevant discussion is in Ref.~\citenum{Martin2003} Ch.~5. The system is fully specified by only two parameters: $r_s = \left( \cfrac{3}{4 \pi \rho} \right)^{1/3} \equiv \mathtt{rs}$ and the shell number. For a momentum eigenvalue $\mathbf{k}$, the shell number is defined by mutually exclusive sets of Fermi surfaces\footnote{Note that the Fermi surface is spherical only in the case of noninteracting fermions. Interaction causes a slight deviation from a perfect sphere.} in $\mathbf{k}$-space which contain the same number of allowed $\mathbf{k}$ states. The simulation cell in this case is defined with periodic boundary conditions (PBC), resulting in a discrete set of allowed $\mathbf{k}$ values. The smallest Fermi surface containing more than one $\mathbf{k}$ state contains 7 $\mathbf{k}$ states which include $\mathbf{k} = \mathbf{0}$ and its 6 nearest neighbors. The next smallest set of Fermi surfaces (``shells'') contain 19 $\mathbf{k}$ and the numbers continue in a sequence of 27, 33, 57, 81, etc.
@ -324,46 +115,14 @@ QMCPACK currently supports only closed-shell systems, meaning that the allowed n
%Key parameters \icode{} sets rs of the problem. condition is used to set the number of particles at the given rs.
The path of the key parameters of the simulation (XPath) are as follows.
\begin{lstcpp}
\begin{code}
simulationcell/parameter/@name='rs'
particleset/group/@size : number of particles per spin
determinantset/@shell
\end{lstcpp}
\end{code}
Shown below is a typical form of the input file.
\begin{lstcpp}
<qmcsystem>
<simulationcell>
<!-- Note 1.
change condition for different number of electrons. Only closed-shell
systems are supported for the moment. The allowed numbers are:
14 38 54 66 114 162 ....
-->
<parameter name="rs" condition="14">5</parameter>
<parameter name="bconds">p p p</parameter>
<parameter name="LR_dim_cutoff">6</parameter>
</simulationcell>
<particleset name="e" random="yes">
<group name="u" size="7">
<parameter name="charge">-1</parameter>
</group>
<group name="d" size="7">
<parameter name="charge">-1</parameter>
</group>
</particleset>
<hamiltonian name="h0" type="generic" target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
</hamiltonian>
<wavefunction name="psi0" target="e">
<!-- Note 2.
change shell according to the parameter/@condition
14(1) 38(2) 54 (3) 66 (4) 114 (5) 162 (6) ...
-->
<determinantset type="electron-gas" shell="1" randomize="true"/>
<jastrow name="Jee" type="Two-Body" function="rpa"/>
</wavefunction>
</qmcsystem>
\end{lstcpp}
\codesrc{3dheg.s000.xml}
\section{Spherical systems}
\subsection{Spherical jellium}
% Specific Example Calculations

228
docs/usrdoc/qmcpack-u-gs.tex
View File

@ -2,9 +2,9 @@
%developer contact info here (Jeongnim? or all of us?)
\section{Obtaining QMCPACK}
QMCPACK is written in C++ and is currently under the New BSD License. Due to its incomplete compliance with the current ISO C++ standard, it is not released in cross-compilable packages. It is instead obtained from its source code repository hosted by Google Code. As described in project website at http://code.google.com/p/qmcpack/, the latest revision of the code can be checked out anonymously using Subversion, with the command
\begin{lstcnsl}
\begin{console}
svn checkout http://qmcpack.googlecode.com/svn/trunk/ qmcpack-read-only
\end{lstcnsl}
\end{console}
where the local directory name \icode{qmcpack-read-only} can be changed as needed. The repository forbids anonymous users (ie. nonmembers of the project) from committing changes to the code. The above command will create a directory \icode{qmcpack-read-only} (topdir) which contains
\begin{verbatim}
qmcpack-read-only/
@ -30,7 +30,7 @@ A compromise must be made between overly many case-by-case examples and a simpli
%Theoretically speaking,
QMCPACK can be used on any *nix machine with C++ compilers that are reasonably recent. Some examples to date are g++ $\geq 4.2$ (GNU) and icc $\geq 10.1$ (Intel). There are other required packages which the compilation process will point out if missing. Listed below are those required specifically by QMCPACK, with the oldest tested versions.
\begin{itemize}
\begin{itemize*}
\item{} CMake $\geq 2.8$
\item{} mpich, mvapich or openmpi
\item{} blas or lapack library (provided as mkl in Intel compilers)
@ -38,28 +38,31 @@ QMCPACK can be used on any *nix machine with C++ compilers that are reasonably r
\item{} fftw library $\geq 3$
\item{} xml2 library
\item{} hdf5 library $\geq 1.8$
\end{itemize}
For library packages, their full development versions (with suffixes -dev or -devel) are needed. No additional environment variables need to be set if the above packages are installed from precompiled binaries.\footnote{Typical examples are rpm or deb packages.} If instead the libraries are locally built, their installation directories must be explicitly set as environment variables, eg. \icode{LIBXML2\_HOME=/usr/local}. See the installation notes in each package for details.
\item{} pygtkglext: OpenGl extensions for Python GTK bindings for the GUI interface.
% This package has some dependencies.
\end{itemize*}
For library packages, their full development versions (with suffixes -dev or -devel) are needed. No additional environment variables need to be set if the above packages are installed from precompiled binaries.\footnote{Typical examples are rpm or deb packages.} If instead the libraries are locally built, their installation directories must be explicitly set as environment variables, eg. \icode{LIBXML2\_HOME=/usr/local}. See the installation notes in each package for details. On many HPC centers, these packages are managed by utilities like module and softenv. Again, see the documentation on each site for details.
% recent [meaning $> 4.2$ at least] GNU
%Compiling the QMCPACK source requires the CMake package.
%For cmake documentation and guides, consult cmake wiki.
Assuming that all of the above packages are properly set up, building QMCPACK proceeds as follows. Change to qmcpack the top directory (name can be different).
\begin{lstcnsl}
\begin{console}
cd qmcpack-read-only
\end{lstcnsl}
We recommend out-of-source compilation by creating a directory for the libraries and binaries that is separate from the source directory. If not in \icnsl{qmcpack-read-only} already, create the build directory (eg. \icnsl{build}) then change to the directory.\footnote{Like the top directory, the build directory name is completely arbitrary and it can be created at any location, even outside \icnsl{qmcpack-read-only}. This is useful if you need more than one build, for testing purposes.}
\begin{lstcnsl}
\end{console}
We recommend out-of-source compilation by creating a directory for the libraries and binaries that is separate from the source directory. If not in \iconsole{qmcpack-read-only} already, create the build directory (eg. \iconsole{build}) then change to the directory.\footnote{Like the top directory, the build directory name is completely arbitrary and it can be created at any location, even outside \iconsole{qmcpack-read-only}. This is useful if you need more than one build, for testing purposes.}
\begin{console}
mkdir build
cd build
\end{lstcnsl}
\end{console}
In the build directory, run cmake to create Makefiles, then build the executable using make.
%(Don't forget ..)
\begin{lstcnsl}
\begin{console}
cmake ..
make
\end{lstcnsl}
If everything goes well, then you should see \icnsl{qmcpack-read-only/build/bin/qmcapp}.
\end{console}
If everything goes well, then you should see \iconsole{qmcpack-read-only/build/bin/qmcapp}.
%The procedure above, creating build directory and running camke in a new directory, is an example. We can further separate the source (development) and build. Let's assume that the QMCPACK topdir is /home/foo/src/qmcpack. Then, one can build multiple executables in different locations by creating new directories and build QMCPACK in each directory.
%\begin{verbatim}
%/home/foo/build/gcc-real
@ -73,153 +76,126 @@ If everything goes well, then you should see \icnsl{qmcpack-read-only/build/bin/
%$cmake /home/foo/src/qmcpack
%$make
%\end{verbatim}
So far, there is no need to change sources or cmake files. \icode{cmake ..} in the above procedure uses \icnsl{..} because the source tree resides in the parent directory. If something did not work, simply remove the directory (eg. \icode{rm -rf build}) and start again.
So far, there is no need to change sources or cmake files. \icode{cmake ..} in the above procedure uses \iconsole{..} because the source tree resides in the parent directory. If something did not work, simply remove the directory (eg. \icode{rm -rf build}) and start again.
Additional configurations must be considered if you need to deviate from the default settings of parallel computing (MPI) and multithreading (OpenMP). See \S{}\ref{ss:compset} for details.
%discuss toolchains here
\subsection{Compiler settings for MPI and OpenMP} \label{ss:compset}
%To run QMCPACK under MPI,
\paragraph{MPI}
Running QMCPACK under MPI requires it to be built with a modified compiler. This usually involves installing an additional \emph{compiler wrapper} package which turns the already working compiler into an MPI-capable one, without having to rebuild the compiler itself.\footnote{MPI implementations of this type are known as \emph{portable implementations}.} Examples of MPI implementations known to work with QMCPACK are OpenMPI, MPICH2, and MVAPICH.
%icc vs. g++
Several environment variables must be set in order to use MPI.
Some environment variables must be set in order to use MPI. By default, the variable CXX is set to the serial compiler (either g++ or icc). Change this to one of the following:
%MPI is automatically enabled if
\begin{itemize*}
\item{} CXX is set to parallel compilers
\begin{itemize*}
\item{} mpicxx, mpic++, cmpic++ (tungsten at NCSA)
\item{} mpCC/mpCC\_r on AIX
\end{itemize*}
\item{} mpi.h is found in standard paths, e.g., /usr/include or /usr/local/include
\begin{itemize*}
\item{} SGI Altix [?]
\end{itemize*}
\item{} mpicxx, mpic++, cmpic++ (tungsten at NCSA)
\item{} mpCC/mpCC\_r on AIX
\end{itemize*}
One of these actions will disable MPI
Next, add the directory containing mpi.h to the include file search path. If the compiler wrappers were installed from binary packages or were locally built with default options, they should already be in one of the directories in the standard search path, eg. /usr/include or /usr/local/include. If you want to keep MPI enabled for other projects but disable it for QMCPACK only, do one of the following:
%s, eg., /usr/include or /usr/local/include
% \begin{itemize*}
% \item{} SGI Altix [?]
% \end{itemize*}
%One of these actions will disable MPI
\begin{itemize*}
\item{} Set \icode{QMC\_MPI} environment to 0, eg. \icode{export QMC\_MPI=0} for bash.
\item{} Modify \icnsl{topdir/CMakeLists.txt}:
\begin{verbatim}
\item{} Modify \iconsole{topdir/CMakeLists.txt}:
\begin{code}
SET(QMC_MPI 0)
\end{verbatim}
\end{code}
\item{} Set \icode{QMC\_MPI} environment to 0, eg. \icode{export QMC\_MPI=0} for bash.
\end{itemize*}
\textbf{Note}: Shell environment variables take precedence over CMake settings.
OpenMP is disabled by default in \icnsl{CMakeLists.txt}. But, OpenMP is automatically enabled if CMake detects that these compilers are being used:
\begin{itemize}
\paragraph{Multithreading}
OpenMP is disabled by default in \iconsole{CMakeLists.txt} but is automatically enabled if CMake detects that these compilers are being used:
\begin{itemize*}
\item{} Intel compilers
\item{} IBM VisualAge compilers
\item{} GNU/OpenMP compilers $>$ 4.2.x on Linux 2.6.x kernels, Mac OS X
\end{itemize}
\end{itemize*}
To enable OpenMP for other compilers, one of the following actions will be needed:
\begin{itemize}
\item{} Set \icode{QMC\_OMP} to 1, eg. \icode{export QMC\_OMP=1} for bash.
\item{} Modify \icnsl{topdir/CMakeLists.txt}:
\begin{verbatim}
\begin{itemize*}
\item{} Modify \iconsole{topdir/CMakeLists.txt}:
\begin{code}
SET(QMC_OMP 1)
\end{verbatim}
\end{itemize}
\end{code}
\item{} Set \icode{QMC\_OMP} to 1, eg. \icode{export QMC\_OMP=1} for bash.
\end{itemize*}
If your machine has multiple cores, there is no need to disable OpenMP. However, make sure to set the environment variables which control OpenMP runs. Especially with MKL, set
\begin{verbatim}
[check this] If your machine has multiple cores, there is no need to disable OpenMP. However, make sure to set the environment variables which control OpenMP runs. Especially with MKL, set
\begin{code}
MKL_NUM_THREADS=1
MKL_SERIAL=YES
\end{verbatim}
\end{code}
so that the blas/lapack calls DO NOT USE threaded version.
Note that the default number of threads on your machine may be set to the number of cores (or CPU units). It is always safe to set the number of threads yourself as
\begin{verbatim}
export OMP\_NUM\_THREADS=1
\end{verbatim}
Note that the default number of threads on your machine may be set to the number of cores (or CPU units). It is always safe to set the number of threads yourself as, for example,
\begin{code}
export OMP_NUM_THREADS=1
\end{code}
More on cmake
cmake environment variables
QMCPACK/cmake uses environment variables to determine compiler-time options.
These are most critical environment variables with default values
\begin{verbatim}
QMC_MPI=1
\end{verbatim}
1/0 to enable/disable MPI
\begin{verbatim}
QMC_OMP=0
\end{verbatim}
1/0 to enable/disable OpenMP
\begin{verbatim}
QMC_COMPLEX=0
\end{verbatim}
1/0 to use complex/real for the wavefunctions
\begin{verbatim}
QMC_BITS=32
\end{verbatim}
32/64 for the OS
Each build should use identical variables. If the working shell has different variables from the previous build environments, cmake/make will rebuild everything which can take time. Out-of-source compilation becomes very useful to build different combinations.
Note that separate executables have to be built for real and complex wavefunctions. One can use real wavefunction for the complex-enabled build but it will be extremely inefficient (4-8 times slower). However, complex wavefunctions cannot be used with real-enabled build.
%More on cmake
%cmake environment variables
In addition to \icode{QMC\_MPI} and \icode{QMC\_OMP}, there are few more environment variables used by QMCPACK and CMake to determine compiler-time options. \icode{QMC\_COMPLEX} (0 or 1) sets wavefunctions to take real or complex values.\footnote{One can use real wavefunction for the complex-enabled build but it will be extremely inefficient (4-8 times slower). However, complex wavefunctions cannot be used with real-enabled build.} \icode{QMC\_BITS} (32 or 64) sets the OS bit size. A change in any one of these variables will cause Make to rebuild everything instead of rebuilding just the modified source files.
%Each build should use identical variables. If the working shell has different variables from the previous build environments, cmake/make will rebuild everything which can take time. Out-of-source compilation becomes very useful to build different combinations.
%Note that separate executables have to be built for real and complex wavefunctions.
%\subsection{Library dependencies}
\subsection{Required utilities and libraries}
In order to install QMCPACK, users have to install several required packages. These packages are included in standard Linux/cygwin distributions or can be downloaded by following the links. If these libraries are installed in standard directories, /usr /usr/local and /sw (Mac), no action is necessary. Alternatively, environment variables XYZ\_HOME should be set. Here, XYZ stands for the name of package; the build utility can locate the libraries and use them.
%\subsection{Required utilities and libraries}
%In order to install QMCPACK, users have to install several required packages. These packages are included in standard Linux/cygwin distributions or can be downloaded by following the links. If these libraries are installed in standard directories, /usr /usr/local and /sw (Mac), no action is necessary. Alternatively, environment variables XYZ\_HOME should be set. Here, XYZ stands for the name of package; the build utility can locate the libraries and use them.
With few exceptions, the build utility cmake will look for XYZ\_HOME/include for the header files and XYZ\_HOME/lib for the library files. When multiple environment variables apply to a library, e.g., blas/lapack, the library is searched according to the listed order.
%With few exceptions, the build utility cmake will look for XYZ\_HOME/include for the header files and XYZ\_HOME/lib for the library files. When multiple environment variables apply to a library, e.g., blas/lapack, the library is searched according to the listed order.
%intel mkl, gsl, boost, mpich/openmpi, openmp, cuda, etc.
\begin{itemize}
\item{} cmake utility source and binary distribution on multiple platforms
\item{} blas/lapack Numerical library MKL\_HOME, LAPACK, ATLAS Alternatives: vendor-provided blas, e.g., ESSL
\item{} hdf5 I/O HDF5\_HOME, HDF\_HOME source and binary distribution on multiple platforms
\item{} libxml2 I/O LIBXML2\_HOME Standard library for Linux distributions
\item{} boost C++ standard libraries BOOST\_HOME Using only the header files. No need to compile anything. Simply download and unpack the package.
\item{} einspline c library for 3D bspline see Using Einspline library
\item{} fftw c library for FFT FFTW\_HOME http://www.fftw.org/
\item{} pygtkglext OpenGl extensions for Python GTK bindings For the GUI interface. This package has some dependencies.
\end{itemize}
%\begin{itemize}
%\item{} cmake utility source and binary distribution on multiple platforms
%\item{} blas/lapack Numerical library MKL\_HOME, LAPACK, ATLAS Alternatives: vendor-provided blas, e.g., ESSL
%\item{} hdf5 I/O HDF5\_HOME, HDF\_HOME source and binary distribution on multiple platforms
%\item{} libxml2 I/O LIBXML2\_HOME Standard library for Linux distributions
%\item{} boost C++ standard libraries BOOST\_HOME Using only the header files. No need to compile anything. Simply download and unpack the package.
%\item{} einspline c library for 3D bspline see Using Einspline library
%\item{} fftw c library for FFT FFTW\_HOME http://www.fftw.org/
%
%\end{itemize}
On many HPC centers, these packages are managed by utilities like module and softenv. Consult the documentations on each site.
\paragraph{Numerical Libraries}
%If lapack/blas or atlas is not in your standard path, do one of the following. ``location'' is where the libraries are located.
If LAPACK/BLAS or ATLAS is not in your library search path, set the environment variables \icode{LAPACK} to \icode{"-L/installdir -llapack -lblas"} and \icode{ATLAS} to\\ \icode{"-L/installdir -llapack -lf77blas -lcblas -latlas"}, changing \iconsole{installdir} appropriately.
Notes on Numerical Libraries: If lapack/blas or atlas is not in your standard paths, do one of the followings. location is where the libraries are located.
For bash users,
\begin{verbatim}
export LAPACK="-L/location -llapack -lblas"
export ATLAS="-L/location -llapack -lf77blas -lcblas -latlas"
\end{verbatim}
For tcsh users,
\begin{verbatim}
setenv LAPACK "-L/location -llapack -lblas"
setenv ATLAS "-L/location -llapack -lf77blas -lcblas -latlas"
\end{verbatim}
%For bash users,
%\begin{verbatim}
% export LAPACK="-L/location -llapack -lblas"
% export ATLAS="-L/location -llapack -lf77blas -lcblas -latlas"
%\end{verbatim}
%For tcsh users,
%\begin{verbatim}
% setenv LAPACK "-L/location -llapack -lblas"
% setenv ATLAS "-L/location -llapack -lf77blas -lcblas -latlas"
%\end{verbatim}
\subsection{Configurations known to work}
\subsubsection{Intel 64 Abe cluster \@ NCSA}
* They are usually installed by the system admins of super computing centers. If not, request the libraries.
* Use standard Linux package management tools, e.g., dpkg, yum, to install them.
Before building
* Use default mpi library: +mvapich2-intel
* Add using SoftEnv: in .soft file or in your login shell
o +intel-mkl
o +hdf5-1.6.5
o +libxml2-2.6.29
* boost is available at /u/ac/jnkim/share/boost
* Set environment variables in your shell
\begin{verbatim}
CXX=mpicxx
CC=icc
LIBXML2_HOME=/usr/local/libxml2-2.6.29
QMC_BITS=64
QMC_OMP=1
\end{verbatim}
\subsubsection{Intel 64 Abe cluster @ NCSA}
Abe uses SoftEnv to activate the configurations for the (otherwise dormant) software installed in it. Create the file .soft in the home directory, containing the following lines:
\begin{console}
@default\\
@boost-1.42.0\\
@teragrid-basic\\
+intel-mkl\\
+libxml2\\
+hdf5-1.8.4
\end{console}
Set the following environment variables:
\begin{console}
CXX=mpicxx\\
CC=icc\\
LIBXML2\_HOME=/usr/local/libxml2-2.6.29\\
QMC\_BITS=64\\
QMC\_OMP=1
\end{console}
\subsubsection{Intel Mac OS X}
Intel Mac OS X
From QMCPACK
[I can try this on my laptop, with MacPorts]
Basic steps
@ -280,7 +256,7 @@ fi
QMCPACK is developed for large-scale high-performance computing systems. We update the status of QMCPACK on the HPC systems the developers have access to and share experiences in dealing with some quirkiness of each system.
In general, the quickest way to build and use QMCPACK is to use a toolchain file for each system. They are available in config directory with the distribution. Current list includes
\begin{itemize}
\begin{itemize*}
\item{} AbeMvapich2.20091104.cmake : abe@ncsa
\item{} AbeMvapich2.cmake
\item{} AbeOpenMPI.cmake
@ -290,8 +266,8 @@ In general, the quickest way to build and use QMCPACK is to use a toolchain file
\item{} KrakenGNU.cmake : kraken@nics, Cray XT
\item{} LinuxIntel.cmake : generic for LINUX using Intel compilers
\item{} LOP6\_xlC\_xlc.cmake : huygens@sara, IBM P6, running LINUX, IBM XL compilers
\item{} LOP6\_xlC\_gcc.cmake: huygens@sara, IBM P6, running LINUX, IBM XLC compiler and GNU C compiler
\end{itemize}
\item{} LOP6\_xlC\_gcc.cmake : huygens@sara, IBM P6, running LINUX, IBM XLC compiler and GNU C compiler
\end{itemize*}
Once a toolchain file (mychain.cmake) is selected,
\begin{verbatim}
cd build

2
docs/usrdoc/qmcpack-u-run.tex
View File

@ -5,7 +5,7 @@ QMCPACK uses MPI/OpenMP hybrid method for the parallelization. The optimal choic
\item{} memory hierarchy.
\end{itemize}
For the common multicore chips of today, using one MPI per node and setting \verb|OMP\_NUM\_THREADS| to the number of cores of a node will work. But, if a walker can be fit into a NUMA node, it is better to use one MPI per NUMA node. For instance, on Cray XT5 with dual hex-core AMD chips, setting \verb|OMP\_NUM\_THREADS=6| will work best. [check this]
For the common multicore chips of today, using one MPI per node and setting \icode{OMP\_NUM\_THREADS} to the number of cores of a node will work. But, if a walker can be fit into a NUMA node, it is better to use one MPI per NUMA node. For instance, on Cray XT5 with dual hex-core AMD chips, setting \icode{OMP\_NUM\_THREADS=6} will work best. [check this]
Setting the number of walkers and samples can be tricky with parallel runs. The basic rule is at least one walker per thread. See Setting samples to learn how to control QMC runs.

38
docs/usrdoc/qmcpack-u.tex
View File

@ -4,35 +4,23 @@
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{fullpage}
%\usepackage{graphicx}
\usepackage[bookmarks=false,colorlinks=true,linkcolor={blue},pdfstartview={XYZ null null 1.22}]{hyperref}
\usepackage[usenames,dvipsnames]{color}
\usepackage[bookmarks=false,colorlinks=true,linkcolor={blue},pdfstartview={XYZ null null 1.22}]{hyperref}
\usepackage{fancyvrb}
%\usepackage{verbatim}
\usepackage{listings}
%\usepackage{listings}
\usepackage{natbib,mdwlist}
\usepackage[tracking=true]{microtype}
%\usepackage[tracking=true]{microtype}
%\DeclareMicrotypeSet*[tracking]{my}
% { encoding = *,
% size = {-small,Large-},
% font = */*/*/tt/* }
\definecolor{codebg}{gray}{0.80}
\SetTracking{encoding=*,shape=tt*}{-20}
\lstloadlanguages{[ISO]C++}
\newcommand{\icode}[1]{\textcolor{OliveGreen}{\texttt{#1}}}
\newcommand{\icnsl}[1]{\textcolor{RoyalBlue}{\texttt{#1}}}
\lstnewenvironment{lstcpp}{\lstset{backgroundcolor=\color{codebg},language=[ISO]C++,basicstyle=\color{OliveGreen}\ttfamily}}{}
%\lstnewenvironment{lstxml}{\lstset{backgroundcolor=\color{codebg},language=XML}}{}
%\lstnewenvironment{lstcmake}{\lstset{backgroundcolor=\color{codebg},language=make}}{}
%\lstnewenvironment{lstcnsl}{\lstset{backgroundcolor=\color{codebg},language=bash,basicstyle=\color{OliveGreen}\tt}}{}
\lstnewenvironment{lstcnsl}{\lstset{backgroundcolor=\color{codebg},basicstyle=\color{OliveGreen}\ttfamily}}{}
%\newenvironment{code}{\color{OliveGreen}\verbatim}{\endverbatim}
%\newenvironment{fpath}{\color{RoyalBlue}\verbatim}{\endverbatim}
%\newenvironment{code}{\fbox\bgroup}{\egroup}
%\makeatletter\newenvironment{code}{%
% \begin{lrbox}{\@tempboxa}\begin{minipage}{0.9\columnwidth}}{\end{minipage}\end{lrbox}%
% \colorbox{OliveGreen}{\usebox{\@tempboxa}}
%}\makeatother
%\definecolor{codebg}{gray}{0.80}
%\newcommand{\icode}[1]{\textcolor{OliveGreen}{\texttt{#1}}}
%\newcommand{\iconsole}[1]{\textcolor{RoyalBlue}{\texttt{#1}}}
\newcommand{\icode}[1]{\textcolor{OliveGreen}{\Verb|#1|}}
\newcommand{\iconsole}[1]{\textcolor{RoyalBlue}{\Verb|#1|}}
\newcommand{\codesrc}[1]{\VerbatimInput[frame=single,rulecolor=\color{black},formatcom=\color{OliveGreen},numbers=left]{#1}}
\newcommand{\consolesrc}[1]{\VerbatimInput[frame=single,rulecolor=\color{black},formatcom=\color{RoyalBlue},numbers=none]{#1}}
\DefineVerbatimEnvironment{code}{Verbatim}{numbers=left,frame=single,rulecolor=\color{black},formatcom=\color{OliveGreen},numbers=none}
\DefineVerbatimEnvironment{console}{Verbatim}{numbers=left,frame=single,rulecolor=\color{black},formatcom=\color{RoyalBlue},numbers=none}
\title{QMCPACK User Documentation}
\author{UIUC Electronic Structure Group}