MPB release notes
From AbInitio
(Difference between revisions)
Revision as of 02:21, 3 April 2014 (edit) Stevenj (Talk  contribs) ← Previous diff 
Current revision (02:22, 3 April 2014) (edit) Stevenj (Talk  contribs) 

Line 12:  Line 12:  
* Support FFTW version 3.x (in addition to FFTW 2.x). Version 3.3 or later is required for MPI parallelism.  * Support FFTW version 3.x (in addition to FFTW 2.x). Version 3.3 or later is required for MPI parallelism.  
* Support for OpenMP parallelism.  * Support for OpenMP parallelism.  
  * Use more accurate subpixel averaging algorithm for interfaces between anisotropic material (see [http://math.mit.edu/~stevenj/papers/papers_abstracts.html#KottkeFa08 Kottke et al, PRE 77, 036611, 2008]).  +  * Use more accurate subpixel averaging algorithm for interfaces between anisotropic materials (see [http://math.mit.edu/~stevenj/papers/papers_abstracts.html#KottkeFa08 Kottke et al, PRE 77, 036611, 2008]). 
* Use more accurate geometry routines in recent libctl versions (improves and speeds subpixel averaging).  * Use more accurate geometry routines in recent libctl versions (improves and speeds subpixel averaging).  
* Support using a different '''k''' origin in <code>findk</code>.  * Support using a different '''k''' origin in <code>findk</code>. 
Current revision
MIT Photonic Bands 
Download 
Release notes 
MPB manual 
Introduction 
Installation 
User Tutorial 
Data Analysis Tutorial 
User Reference 
Developer Information 
Acknowledgements 
License and Copyright 
Here, we describe what has changed between releases of the MIT Photonic Bands package. You can also refer to the NEWS
file in the MPB package (or the ChangeLog
file for a more detailed listing).

[edit]
MPB 1.5
April 2, 2014
 MPB now also installs a library, for use from within Meep 1.2 or later.
 Support Guile 2.x.
 Support FFTW version 3.x (in addition to FFTW 2.x). Version 3.3 or later is required for MPI parallelism.
 Support for OpenMP parallelism.
 Use more accurate subpixel averaging algorithm for interfaces between anisotropic materials (see Kottke et al, PRE 77, 036611, 2008).
 Use more accurate geometry routines in recent libctl versions (improves and speeds subpixel averaging).
 Support using a different k origin in
findk
.  Added
epsilonfunc
wrapper formaterialfunc
(similar to Meep).  Added
compute1groupvelocity
to compute group velocity of a single band.  Added
kinterpolateuniform
function, to interpolate with roughly uniform spacing in reciprocal space.  Added
optimizegridsize!
function to round the grid size to a size that can be handled more efficiently. (Only affects resolution, not lattice vectors.)  Allow user to set
filenameprefix
to false to disable its use.  Added
mpbdata P <angle>
option to change the phase angle of the output.  Added
computefielddivergence
and routines to get and output the bound charge density:getchargedensity
andoutputchargedensity
. 
resolution
is now an arbitrary real number, not just an integer, although of course MPB eventually computes an integer grid size.  Support HDF5 1.8.
 Fix recurring non positivedefinite matrix in potrf errors that were arising due to roundoff errors preventing matrix inversion.
 Fix
outputatkpoint
to avoid sensitivity to roundoff errors.  Bug fix in parallel HDF5 support: HDF5 compiled for parallel I/O (MPI) now works.
 Bug fix in
fieldmap!
, thanks to Karen Lee for the bug report.  Bug fix for
firstbrillouinzonek
, thanks to Mischa Megens.  Bug fix:
getfield
andcomputefieldintegral
now use fields with the same phase as the outputted fields. Thanks to Jim West for the bug report.  Miscellaneous bug fixes.
[edit]
MPB 1.4.2
March 3, 2003
 Interactive prompt is now "
mpb>
" not "guile>
".  Output "
freqs:
" line lists headings as "k1, k2, k3
" instead of "kx, ky, kz
" since they are in reciprocallattice, not cartesian, coordinates. Thanks to Theis Peter Hanson for the suggestion.  Bug fix in
findk
for nonorthogonal lattices; thanks to Suxia (Susan) Yang for tracking down this bug.  Fixed SunOS problem where k vectors along
nosize
dimensions failed; thanks to Benjamin Cowan for the bug report.  Fixed
findk
to work forbandmin
> 1; thanks to M. Povinelli for the bug report.  Fixed
findk
to work for thunk band functions (which take no arguments and are called only once instead of perband).
[edit]
MPB 1.4.1
September 16, 2002
 Fixed
NaN
in field normalization when basis determinant was negative. Thanks to Rumen Iliew for the bug report.  Fixed compatibility problems with versions of Guile prior to 1.4. Thanks to Cazimir G. Bostan for bug reports.
 Don't resize lattice basis for
gridsize
== 1 unlessnosize
was explicitly specified; thanks to Tairan Wang for the suggestion.
[edit]
MPB 1.4
September 12, 2002
 New
findk
routine to find k as a function of frequency, instead of viceversa.  The Great Field Renormalization: all fields are now normalized to have unit *integral* of their energy density (instead of unit sum over the grid points), which is much more useful e.g. for perturbation theory. (See field normalization section of manual.)
 You can now save fields in Scheme variables to perform computations combining different fields. Example routines, e.g. to output the Poynting vector, are included.
 Functions to export (and import) the raw eigenvectors (planewave amplitudes), as well as to compute dot products of eigenvectors from different kpoints (e.g. for detecting band crossings).

allownegativeepsilon
function to enable negativedielectric support.  Added
examples/dos.scm
to compute density of states via simple Gaussian histogram, suggested by Xavier Gonze and Doug Allan.  Bug fix: allow real offdiagonal epsilon elements without requiring
withhermitianeps
. Thanks to Doug Allan for the bug report.  Eliminated floatingpoint error on Alpha for homogeneous structure. Thanks to F. LopezTejeira for the bug report.
 Added man page for
mpbsplit
.  Minor installation fixes.
[edit]
MPB 1.3
March 10, 2002
 You can now specify the grid size via the
resolution
input variable, instead of viagridsize
. In this case, you make e.g. a 2d simulation by creating a lattice with sizenosize
in one dimension. The old syntax is still supported, but the new style is encouraged (all examples have been updated to the new style).  New functions to retrieve fields, dielectric functions, etcetera at any point, interpolated from the grid if necessary; see the
get*point
functions in the reference section.  New
computefieldintegral
function, analogous tocomputeenergyintegral
; thanks to Marin Soljacic for the suggestion.  Support Scheme complex numbers where appropriate (e.g. in
epsilonoffdiag
or in the new field integration functions).  Got rid of
NaN
when computing the (undefined) group velocity for zerofrequency states at the Gamma point; arbitrarily return zero here instead. Thanks to Dmitry N. Chigrin for reporting floatingpoint exceptions on Alphas.  Fixed compilation failure for Fortran compilers that use all upper case; thanks to Steve Lantz of Cornell.
 Added "Fun with Fortran" section to installation manual describing common Fortran pitfalls; thanks to Steve Lantz for the suggestion.
 Improved BLAS/LAPACK detection; new
withblas
andwithlapack
options to specify these libraries manually.  Shortened
withhermitianepsilon
configure option towithhermitianeps
.  The dataanalysis tutorial is now consistent with h5topng 1.7.
 Use new API from libctl 2.0.
[edit]
MPB 1.2.2
December 7, 2001
 Fixed bug that caused erroneous/failed convergence when
EVENY
/ODDY
constraints were used in three dimensions. Thanks to Rumen Iliew for the bug report.  Added convenience functions
runyeven
,runyodd
,runyevenzeven
, ...
[edit]
MPB 1.2.1
November 20, 2001
 Fixed serious crashing bug in 1.2; thanks to Karl Koch for the bug report.
[edit]
MPB 1.2
November 15, 2001
 Added new yparity computation and constraints. See the new
runparity
function, which allows you to simultaneously specify the parity through the y=0 and z=0 planes, for symmetric structures. See also thedisplayyparities
function.  z parity is no longer computed by default; see the new
displayyparities
anddisplayzparities
functions to pass to(run)
.  Return moreaccurate average epsilon, fill factor, and scalar epsilon values (eigenfrequencies are not affected). Thanks to Mischa Megens for bugging me.
 Now outputs D and H in consistent units (previously, D and E were multiplied by a factor of frequency). Thanks to Michelle Povinelli for worrying.

epsilon.h5
file now includes extra datasets for all components of the effective dielectric tensor. (This feature is not yet supported if youconfigure withinvsymmetry withhermitianepsilon
.) 
runpolarization
is replaced byrunparity
, andruneven
/runodd
are deprecated in favor ofrunzeven
/runzodd
.runte
/runtm
are now equivalent torunzeven
/runzodd
when invoked for 3d systems.  Noted new
basissize
property ofgeometrylattice
, from libctl 1.5. This makes it easier to use conventional units in the fcc lattice.  Groupvelocity computation no longer silently invalidates fields that have been loaded with
getdfield
, etcetera. Thanks to Marin Soljacic for the bug report.  The configure script now checks that
guile
is in the$PATH
. Thanks to Bing Li and Giridhar Malalahalli for their bug reports.  Rotated the W and K points of the diamondlattice example so that they are oriented similarly to those in the Photonic Crystals book by Joannopoulos et al. (eigenfrequencies are not affected). Thanks to Robert Sheldon for pointing out that this was confusing.
 Added
honeyrods.ctl
example file: a 2d honeycomb lattice of rods.  Added
linedefect.ctl
example file: a linedefect waveguide in a 2d triangular lattice of dielectric rods, formed by a missing row of rods.
[edit]
MPB 1.1.1
July 4, 2001
 Fixed bug in Hfield output that caused subtly incorrect Hfield files (only) for 3d problems when NOT using
mpbi
.  Fixed bug that caused
mpbi
to output incorrect results for 1d problems (e.g. outputted dielectric functions with zeros).  Changed default eigensolver tolerance from 1e4 to 1e7.
 Added
retrievegap
convenience function to return the gap between two specified bands.  Fixed typo that prevented compilation of MPI (parallel) version.
 C compiler flags
O3
are no longer used by default, since they don't work with some compilers; most of the performance depends upon the BLAS and FFTW anyway. (Users wishing greater optimization can set theCFLAGS
environment variable.) Thanks to Giridhar Malalahalli for the bug report.
[edit]
MPB 1.1
May 6, 2001
 Added
computeenergyintegral
function to make it easier to compute arbitrary fieldenergy integrals for perturbation theory; thanks to Marin Soljacic for the suggestion.  Fixed bug in
outputfield
routines for the case of a nonzero k_{z} component, that caused the fields to be multiplied by an exp(ikx) phase with a k in the wrong direction. Thanks to Jesper Riishede for the bug report.
[edit]
MPB 1.0
February 23, 2001
 At long last, support for distributedmemory parallel machines with MPI. The computation time (and memory usage) can often improve nearly linearly with the number of processors. Thanks to Clarendon Photonics for funding this work.
 Also added
mpbsplit
script to parallelize in a simpler way, without MPI, on e.g. SMP machines, by dividing up the list of kpoints among a number of serial mpb processes.  Fixed bug in
mpbi
where artifacts could be introduced in 3d field and dielectricfunction output files. (This only affected the output files, not the frequency eigenvalues, etcetera.) Thanks to Michelle Povinelli for the bug report.  Added new
materialfunction
material type, so that you can now specify that the dielectric tensor be an arbitrary function of position. Thanks to Peter Bermel for needing this.  If MPB is configured with the flag
withhermitianepsilon
, then complexhermitian dielectric tensors (corresponding to magnetic materials, which break timereversal symmetry) are supported. Thanks to Shanhui Fan for pestering me about this.  Eliminated
outputcopies
input variable; if you want to visualize multiple unit cells, you should usempbdata
.  Added new "
nothing
" material that punches a hole through other objects to the background. (This is distinct fromdefaultmaterial
whenepsiloninputfile
is used, or forcomputeenergyinobjects
.)  Fixed inability of MPB 0.13 to run under an old version (1.2) of Guile.
 Now gives an error if kpoint or dielectric tensor is incompatible with runte/runtm, or if the dielectric tensor is not positivedefinite.
 Default to vendor
cc
instead ofgcc
, so that C and Fortran compilers are in sync. (We default to the vendorf77
because it was probably used to compile LAPACK/BLAS, and Fortran libraries are picky.)  The manual now cites our recent publication on the methods behind MPB.
 Bug fix in
computeenergyinobjectlist
for nonorthogonal lattices.  Bug fix in
combinebandfunctions
and other functions of band functions, which did not handle functions of no arguments ("thunks") correctly (crashing with an error message). Thanks to Michelle Povinelli for the bug report.  Fixed a floatingpoint sensitivity bug in
mpbdata
that could cause a crash on the Alpha; thanks to Dominique Caron for the bug report and debugging information.
[edit]
MPB 0.13
January 7, 2001
 Can now take advantage of inversion symmetry in the geometry, gaining at least a factor of two in speed and a factor of two in memory. To use this, you configure MPB with
withinvsymmetry
; the resulting executable is installed as "mpbi
" and only supports inversion symmetry, so you will usually want to install the ordinary MPB as well.  Added new
eigensolverblocksize
input variable, so that MPB can optionally solve for only a few bands at a time instead of all at once, reducing memory requirements and often increasing speed.  Improved handling of the singular (zerofrequency) solutions at the Gamma (k=0) point. This k point should no longer converge slowly (or cause additional problems in the targeted eigensolver).
 Manual updates: please see new referencing suggestions; expanded table of contents; we now use more conventional units in diamond/fcc example.
 You can now pass a "thunk" (function of no arguments) to
run
, and it will be evaluated once per kpoint (instead of once per band per kpoint as for ordinary band functions). 
computefieldenergy
function now also returns the fraction of the energy in the various field components. Thanks to Karl Koch for the suggestion.  The
filenameprefix
variable is now read each time an output function is called, instead of once per(run)
, so it can be changed frequently if desired. Thanks to Karl Koch for the suggestion.  Added
firstbrillouinzone
function to transform an arbitrary kpoint into an equivalent point in the first Brillouin zone. Thanks to Payam Rabiei for the suggestion.  In
mpbdata
, the center of the output cell is now always identical to the origin of the coordinate system. Thanks to Michelle Povinelli for pointing out this deficiency.  Used improved sphericalquadrature formula in computing the effective dielectric tensor in 3d; this should increase accuracy somewhat at lower grid resolutions. Thanks to Doug Allan for helpful discussions.
[edit]
MPB 0.12
July 9, 2000
 Added
fix*fieldphase
functions to allow a deterministic phase in the output fields, thanks to a suggestion by Doug Allan.  Added groupvelocity calculation functions (
displaygroupvelocities
, etcetera).  Added
e x,y,z
option tompbdata
so that you can now specify an orientation of the output cell (e.g. to make the first axis the 111 direction of an fcc crystal).  Added
(index n)
substitute forepsilon
property of dielectrics, equivalent to(epsilon (* n n))
.  Documented new libgeom features:
cone
geometric object, coordinate conversion functions (reciprocal>lattice
,lattice>cartesian
, etc.), and vector/matrix rotation. 
computefieldenergy
now returns the total, unnormalized energy in the corresponding field; combined withcomputeenergyinobjects
, this makes it easy to do some perturbation theory and related calculations.  Eigensolver improvements. Periodic reorthogonalization and renormalization to combat some numerical problems. New lineminimization code, included with permission from MINPACK2 by Jorge J. Moré.
 Fixed breaking of 90degree rotational symmetrybreaking by the mesh in 2d; thanks to Jim West and Doug Allan of Corning for the bug report. (In general, some symmetrybreaking by the discretization seems hard to avoid, however.)
 Fixed bug in field output routines that could cause crashes for grid sizes not a multiple of 4.
 Bug fix in dielectric function construction for 2d systems: we now use the xy plane at z=0 as documented, instead of z=0.5.
[edit]
MPB 0.11
February 12, 2000

configure
script can now detect and link ATLAS 3.0 accelerated BLAS.  Added
bandrangedata
output variable.  Running
mpbdata
multiple times on the same file now replaces the results of the previous run, instead of appendingnew2
,new3
, etcetera.  Fixed bug in
runeven
/runodd
that could seriously slow or even prevent eigensolver convergence. Thanks to Payam Rabiei for the bug report.  Fixed compilation
withouthdf5
, or when HDF5 is not found. Thanks to Rajesh Rengarajan for the bug report.
[edit]
MPB 0.10
January 28, 2000
 Added
mpbdata
utility for postprocessing data (e.g. for unskewing nonorthogonal lattices). See the new dataanalysis tutorial orman mpbdata
for more information.  Added new dataanalysis tutorial to the manual, describing how to analyze and visualize the results of two sample calculations.
 Added support for a new material type,
dielectricanisotropic
, so that you can specify arbitrary real/symmetric dielectric tensors.  Added new
outputatkpoint
function to make it easier to output fields only at a single kpoint in a bandstructure calculation.  When outputting fields, output all field components (x, y, z, and real and imaginary parts) to a single HDF5 file. Also include info on the lattice and kpoint vectors to facilitate postprocessing.
 Added new subsection to the installation manual describing some generic installation path issues on Unix that were confusing people.
 Use
CPPFLAGS
environment variable instead of the lessstandardINCLUDES
to passI
flags to the configure script (for header files in nonstandard locations).  Added
diamond.ctl
example file for a 3d diamond (fcc) lattice of spheres.  Added (brief) mpb
man
page.  Fixed zparity output and
runeven
/odd
functions for 2d grids.  Fixed bug in
outputdpwrinobjects
. Thanks to Mihai Ibanescu for the bug report.  Compilation fixes. We need to set
SHELL
in theMakefile
formake
on some systems. Also added rule to insurectlio.h
is created beforemain.c
is compiled. Thanks to Christoph Becher for the bug reports.
[edit]
MPB 0.9.1
January 7, 2000
 Fixed eigensolver bug where special handling of Gamma (k=0) point could screw up convergence for subsequent kpoints, causing incorrect results.
 Fixed behavior of
filenameprefix
input variable; thanks to Karl Koch for the bug report
[edit]
MPB 0.9
January 2, 2000
 Added
runeven
andrunodd
functions, so you can now compute only even/odd states (with respect to a z=0 mirror plane) in systems with sufficient symmetry. See also the newzparity
output variable.  Added
epsiloninputfile
variable, so that you can now read an arbitrary dielectric function from a file.  Field file names now include the polarization (e.g. "
.tm
").  Some optimizations in the eigensolver.
 Some documentation improvements; thanks to Edmond Chow for his comments.

configure
should work even when there is no Fortran compiler on your system (assuming your BLAS, etc., libraries work without Fortran libs). Thanks to Antti Renko for the bug report.  Fixed problems detecting BLAS and LAPACK shared libraries in
configure
. Thanks to Karri Varris for the bug report.  Fixed trailing spaces in
sed
command, which were breakingmake install
on some systems. Thanks to Ron Chase for the bug report.
[edit]
MPB 0.8.1
November 22, 1999
 Added
outputhfieldx
,outputdfieldy
, etcetera, functions for outputting only specific field components.  Sped up HDF5 field output routines.
 Added
outputcopies
variable to set the number of periods output by the band output functions.
[edit]
MPB 0.8
November 19, 1999
 Initial public release.