calclens
Curved-sky grAvitational Lensing for Cosmological Light conE simulatioNS
Copyright (C) 2009-2016 Matthew R. Becker
Released under GNU GPL v3 - see COPYING and AUTHORS for details.
CALCLENS stands for Curved-sky grAvitational Lensing for Cosmological Light conE simulatioNS. It is a curved-sky multiple-plane ray tracing code for generating weak gravitational lensing shear fields from light cone simulations. Currently only flat LCDM models are supported.
License and Conditions of Use
This code is released publicly under GNU GPL v3 (see COPYING for details). The standard GNU disclaimer (which most definitely applies here!) is
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
If CALCLENS is used for any scientific work, please cite the paper describing the code
Becker 2013, MNRAS, 435, 115 [arXiv:astro-ph/1210.3069]
and also post your own work (and any updates) to the arXiv (http://arxiv.org/).
CALCLENS is a BYOLC code, which stands for Bring Your Own Light
Cone. Utilities for reading light cone formats (see lightconeio.c for
how to write your own) are included below.
Compilation
CALCLENS requires the following widely available public software libraries:
GSL - GNU Scientific Library (http://www.gnu.org/software/gsl/)
CFITSIO - for read and writing FITS files (http://heasarc.gsfc.nasa.gov/fitsio/)
HDF5 v1.8 - for the indexed light cone format (http://www.hdfgroup.org/HDF5/)
FFTW3 - for FFTs used by SHTs (http://www.fftw.org/)
Other requirements include:
- a 64 bit system - the types long and double in C must both be 8 bytes
- a Unix/Linux-like environment (It has not been tested on Mac OSX.)
Once these libraries are installed (or if you have them already) and requirements are met, then you just have to edit the Makefile to match your configuration. For Gadget-2 users, this should be very intuitive. Here is an example for the KIPAC Orange cluster
ifeq ($(COMP),"orange")
CC = mpicc
OPTIMIZE = -g -O3
GSLI = -I/afs/slac.stanford.edu/g/ki/software/gsl/1.15/include
GSLL = -L/afs/slac.stanford.edu/g/ki/software/gsl/1.15/lib
FFTWI = -I/afs/slac.stanford.edu/g/ki/software/fftw/3.3/include
FFTWL = -L/afs/slac.stanford.edu/g/ki/software/fftw/3.3/lib
HDF5I = -I/afs/slac.stanford.edu/g/ki/software/hdf5/1.8.8/include
HDF5L = -L/afs/slac.stanford.edu/g/ki/software/hdf5/1.8.8/lib
FITSI = -I/afs/slac.stanford.edu/g/ki/software/cfitsio/3.29/include
FITSL = -L/afs/slac.stanford.edu/g/ki/software/cfitsio/3.29/lib
EXTRACFLAGS =
EXTRACLIB =
endif
Then simply type "make" in the source directory and you should get an executable named "raytrace" for your enjoyment.
Running an MPI code
CALCLENS is an MPI-1 parallel code. See the local user guide (or ask a friend) on how to run MPI code on your local compute cluster. Typically one needs to make a job script and the include commands like
mpirun <path to calclens> <path to config file>
to get this work. See the Ray Tracing below for how to actually run the code.
Ray Tracing
Ray tracing a light cone involves two steps. First, the light cone must be rewritten into the HDF5 index light cone format. Then it can be ray traced.
-
Making the indexed light cone files.
Code to make lens planes is in the lensplanes directory. You will need to setup the configuration file properly as described in Configuration below.
Making lens planes is a purely serial operation and requires a lot of memory (though you do not need to have the entire light cone in memory). The best way to make lens planes is to run an MPI job with as many tasks as you have cores per node. Task 0 will make the lens planes and then at the end all of the tasks read the data to make maps of the matter density for error checking. This can take a while depending on how much data you have, but only has to be done once.
-
Ray Tracing
The ray tracing code is included in the main diretcory. You will need to setup your configuration file properly (see Configuration) and then run the code as described above.
I would recommend, depending on resolution, at least 128 cores for 220 deg^2 patch and then small increases after that as the area gets larger. Note that the SHT steps are the same cost in CPU hours regardless of area, but that for large enough areas the MG steps dominate the running time.
Configuration
CALCLENS uses a configuration file to set most options for running the code. Options not set in the configuration file, but in the Makefile, include
BORNAPPRX - ray trace with born approximation
OUTPUTRAYDEFLECTIONS - output ray deflections in ray output
OUTPUTPHI - output lensing potential in ray output
USE_FITS_RAYOUT - set to use fits for writing rays, otherwise will
use a pure binary format
USE_FULLSKY_PARTDIST - set to tell the code to use a full-sky
particle distribution in the SHT step, but only
use the specified area for the MG step
SHTONLY - set to force the code to use SHTs only
If any of these options are changed, the code must be recompiled.
See raytrace.cfg for an example typical configuration file. Its basic structure is a set of tag-value pairs like this
OmegaM 0.27
The tag names are not case sensitive. All units are assumed to be in comoving Mpc/h, Msun/h, MB, degrees or seconds. Comments are denoted by '#', can appear anywhere on a line except between a tag and its value (i.e. "OmegaM #this is a comment 0.25" is not allowed), and are strongly encouraged.
The options are as follows.
WallTimeLimit - time limit for code in seconds
WallTimeBetweenRestart - time between writing of restart files
OmegaM - matter density in units of critical at z = 0
maxComvDistance - maximum comoving distance to end of light cone
NumLensPlanes - number of lens planes
Note that maxComvDistance and NumLensPlanes should be set so that each lens plane is approximately 25 - 30 Mpc/h.
LensPlanePath - path to lens planes
LensPlaneName - base name of lens plane
When writing lens planes, this path and base name are used to construct the lens plane files like this
<LensPlanePath>/<LensPlaneName>XXXX.h5
where XXXX is the lens plane number (i.e. 0010 for plane 10).
OutputPath - path to directory for code output
RayOutputName - base name of ray outputs
Comment out RayOutputName (by adding a '#' in front of it) to prevent rays from being written to disk. These outputs can be many TB for most calculations so this is usually not recommended.
NumRayOutputFiles - # of files to output rays into
NumFilesIOInParallel - # of files which do I/O at the same time
NumFilesIOInParallel must be less than both NumRayOutputFiles and NumGalOutputFiles. The ray outputs are written to disk like this
<OutputPath>/<RayOutputName>XXXX.YYYY
where XXXX is the lens plane number (from 0 to NumLensPlanes-1) and YYYY is a file index from 0 to NumFilesIOInParallel-1. Note that the code will write an additional ray data output with lens plane number NumLensPlanes which has the rays at the very edge of the light cone (i.e. at maxComvDistance). The rest of the ray outputs are at comoving distance
x*dC + dC/2.0
where x is the lens plane number and dC is the lens plane width given by (maxComvDistance/NumLensPlanes).
The rays, ray tracing area, and Poisson solver are controlled by
bundleOrder - HEALPix order for bundle cells (usually 6 or 7)
rayOrder - HELAPix order for rays (somewhere between 14-16)
minRa - minimum ra for ray tracing area in degrees
maxRa - maximum ray for ray tracing area in degrees
minDec - min dec for ray tracing area in degrees
maxDec - max dec for ray tracing area in degrees
HEALPixRingWeightPath - path to data directory from public HELAPix
SHTOrder - HEALPix order of SHT
ComvSmoothingScale - a comoving smoothing scale (usually ~ a few
N-body softening lengths)
Optionally one can set the parameter
MGConvFact - ratio of residual norm to truncation error on finest
MG level
to control the convergence of the MG code. The code has built in defaults (0.1) so changing this parameter is not recommended.
CALCLENS uses the quadrature weights from the public HEALPix package is HEALPixRingWeightPath is specified. Note that if you make minRa greater than maxRa, then CALCLENS will wrap the domain around the sphere in the other direction (i.e. minRa = 360 and maxRa = 5 will set the domain to go from -5 to +5 degrees).
CALCLENS optionally does a grid search for the lensed images of galaxies placed in the light cone. The source galaxies must be in FITS binary tables in extension 1 with tags px, py and pz which have the comoving location of the galaxy in the light cone in Mpc/h. An ID tag is recommended but is not currently used by the code. To tell CALCLENS about the galaxies, set the following options
GalsFileList - file containing list of source galaxy files
GalOutputName - base output name for galaxy images
NumGalOutputFiles - number of output files for galaxy images
The galaxy output files are formatted like the ray output files with
<OutputPath>/<GalOutputName>XXXX.YYYY
where XXXX is the lens plane number (from 0 to NumLensPlanes-1) and YYYY is a file index from 0 to NumFilesIOInParallel-1.
The galaxy output files are FITS binary tables with the following tags
index - index which denotes which source galaxy this image is for
(see below)
ra - location of image in light in decimal degrees
dec - location of image in light cone in decimal degrees
A00 - 00-component of the lensing Jacobian at the image
A01 - 01-component of the lensing Jacobian at the image
A10 - 10-component of the lensing Jacobian at the image
A11 - 11-component of the lensing Jacobian at the image
The lensing Jacobian has its 0-basis vector along the ra-direction and the 1-basis vector along the dec-direction. The index tag contains a specially constructed source galaxy index which allows one to use multiple source galaxy files easily. It is made in the code like this
index = fileNum + NumGalFiles*<location in file>
where fileNum is the zero-indexed position of the galaxy source file in the GalsFileList, NumGalFiles is the total number of files in the GalsFileList, and is the zero-indexed location of the source galaxy in the fileNum-th source galaxy file. For C programmers this construction should be quite familiar. By knowing how many source galaxy files were input into the code and the index, one can extract the source galaxy file number and location like this
fileNum = index mod NumGalFiles
<location in file> = (index - fileNum)/NumGalFiles
CALCLENS can make lens planes as described above. For making lens planes, the following options must be set.
OutputPath - used to output information about lens planes
NumFilesIOInParallel - controlls I/O loading
OmegaM - see above
maxComvDistance - maximum distance to make planes
NumLensPlanes - number of lens planes
LensPlanePath - path to output lens planes
LensPlaneName - base name for lens planes
rayOrder - see above, not directly used
bundleOrder - see above, not directly used
LightConeFileList - file containing paths to light cone files
LightConeFileType - I/O type for light cone file - see
lightconeio.c for examples
LightConeOriginX - X-origin of light in Mpc/h
LightConeOriginY - Y-origin of light in Mpc/h
LightConeOriginZ - Z-origin of light in Mpc/h
LensPlaneOrder - HEALPix order of cells used to spatially index
the light cone (usually between 3 and 5)
memBuffSizeInMB - size of buffer used in memory to sort particles
(set as big as possible)
MaxNumLensPlaneInMem - if particles for more than this many lens
planes are in mem, all planes will be written to disk (set
to the number of lens planes usually)
LightConePartChunkFactor - memBuffSizeInMB/LightConePartChunkFactor
extra mem is allocated to each plane as needed when sorting
particles (set around 150 or so)
partMass - particle mass in Msun/h for light cone formats which need it
(also used as mass of point mass or NFW mass for those tests)
MassConvFact - conversion factor used to make masses in light cone
into Msun/h units
LengthConvFact - conversion factor used to convert distance units
to Mpc/h
VelocityConvFact - conversion factor to get velocities into km/s
The velocities are not used by the code, but are included so that indexed light cone format can be used more widely.