MPB release notes

From AbInitio

(Difference between revisions)
Jump to: navigation, search
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>find-k</code>. * Support using a different '''k''' origin in <code>find-k</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).

Contents


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 find-k.
  • Added epsilon-func wrapper for material-func (similar to Meep).
  • Added compute-1-group-velocity to compute group velocity of a single band.
  • Added kinterpolate-uniform function, to interpolate with roughly uniform spacing in reciprocal space.
  • Added optimize-grid-size! 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 filename-prefix to false to disable its use.
  • Added mpb-data -P <angle> option to change the phase angle of the output.
  • Added compute-field-divergence and routines to get and output the bound charge density: get-charge-density and output-charge-density.
  • 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 positive-definite matrix in potrf errors that were arising due to roundoff errors preventing matrix inversion.
  • Fix output-at-kpoint to avoid sensitivity to roundoff errors.
  • Bug fix in parallel HDF5 support: HDF5 compiled for parallel I/O (MPI) now works.
  • Bug fix in field-map!, thanks to Karen Lee for the bug report.
  • Bug fix for first-brillouin-zone-k, thanks to Mischa Megens.
  • Bug fix: get-field and compute-field-integral now use fields with the same phase as the outputted fields. Thanks to Jim West for the bug report.
  • Miscellaneous bug fixes.

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 reciprocal-lattice, not cartesian, coordinates. Thanks to Theis Peter Hanson for the suggestion.
  • Bug fix in find-k for non-orthogonal lattices; thanks to Suxia (Susan) Yang for tracking down this bug.
  • Fixed SunOS problem where k vectors along no-size dimensions failed; thanks to Benjamin Cowan for the bug report.
  • Fixed find-k to work for band-min > 1; thanks to M. Povinelli for the bug report.
  • Fixed find-k to work for thunk band functions (which take no arguments and are called only once instead of per-band).

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 grid-size == 1 unless no-size was explicitly specified; thanks to Tairan Wang for the suggestion.

MPB 1.4

September 12, 2002

  • New find-k routine to find k as a function of frequency, instead of vice-versa.
  • 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 k-points (e.g. for detecting band crossings).
  • allow-negative-epsilon function to enable negative-dielectric 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 --with-hermitian-eps. Thanks to Doug Allan for the bug report.
  • Eliminated floating-point error on Alpha for homogeneous structure. Thanks to F. Lopez-Tejeira for the bug report.
  • Added man page for mpb-split.
  • Minor installation fixes.

MPB 1.3

March 10, 2002

  • You can now specify the grid size via the resolution input variable, instead of via grid-size. In this case, you make e.g. a 2d simulation by creating a lattice with size no-size 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 compute-field-integral function, analogous to compute-energy-integral; thanks to Marin Soljacic for the suggestion.
  • Support Scheme complex numbers where appropriate (e.g. in epsilon-offdiag or in the new field integration functions).
  • Got rid of NaN when computing the (undefined) group velocity for zero-frequency states at the Gamma point; arbitrarily return zero here instead. Thanks to Dmitry N. Chigrin for reporting floating-point 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 --with-blas and --with-lapack options to specify these libraries manually.
  • Shortened --with-hermitian-epsilon configure option to --with-hermitian-eps.
  • The data-analysis tutorial is now consistent with h5topng 1.7.
  • Use new API from libctl 2.0.

MPB 1.2.2

December 7, 2001

  • Fixed bug that caused erroneous/failed convergence when EVEN-Y/ODD-Y constraints were used in three dimensions. Thanks to Rumen Iliew for the bug report.
  • Added convenience functions run-yeven, run-yodd, run-yeven-zeven, ...

MPB 1.2.1

November 20, 2001

  • Fixed serious crashing bug in 1.2; thanks to Karl Koch for the bug report.

MPB 1.2

November 15, 2001

  • Added new y-parity computation and constraints. See the new run-parity function, which allows you to simultaneously specify the parity through the y=0 and z=0 planes, for symmetric structures. See also the display-yparities function.
  • z parity is no longer computed by default; see the new display-yparities and display-zparities functions to pass to (run).
  • Return more-accurate 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 you configure --with-inv-symmetry --with-hermitian-epsilon.)
  • run-polarization is replaced by run-parity, and run-even/run-odd are deprecated in favor of run-zeven/run-zodd. run-te/run-tm are now equivalent to run-zeven/run-zodd when invoked for 3d systems.
  • Noted new basis-size property of geometry-lattice, from libctl 1.5. This makes it easier to use conventional units in the fcc lattice.
  • Group-velocity computation no longer silently invalidates fields that have been loaded with get-dfield, 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 diamond-lattice 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 honey-rods.ctl example file: a 2d honeycomb lattice of rods.
  • Added line-defect.ctl example file: a line-defect waveguide in a 2d triangular lattice of dielectric rods, formed by a missing row of rods.

MPB 1.1.1

July 4, 2001

  • Fixed bug in H-field output that caused subtly incorrect H-field 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 1e-4 to 1e-7.
  • Added retrieve-gap 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 the CFLAGS environment variable.) Thanks to Giridhar Malalahalli for the bug report.

MPB 1.1

May 6, 2001

  • Added compute-energy-integral function to make it easier to compute arbitrary field-energy integrals for perturbation theory; thanks to Marin Soljacic for the suggestion.
  • Fixed bug in output-field routines for the case of a nonzero kz 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.

MPB 1.0

February 23, 2001

  • At long last, support for distributed-memory 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 mpb-split script to parallelize in a simpler way, without MPI, on e.g. SMP machines, by dividing up the list of k-points among a number of serial mpb processes.
  • Fixed bug in mpbi where artifacts could be introduced in 3d field and dielectric-function output files. (This only affected the output files, not the frequency eigenvalues, etcetera.) Thanks to Michelle Povinelli for the bug report.
  • Added new material-function 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 --with-hermitian-epsilon, then complex-hermitian dielectric tensors (corresponding to magnetic materials, which break time-reversal symmetry) are supported. Thanks to Shanhui Fan for pestering me about this.
  • Eliminated output-copies input variable; if you want to visualize multiple unit cells, you should use mpb-data.
  • Added new "nothing" material that punches a hole through other objects to the background. (This is distinct from default-material when epsilon-input-file is used, or for compute-energy-in-objects.)
  • Fixed inability of MPB 0.13 to run under an old version (1.2) of Guile.
  • Now gives an error if k-point or dielectric tensor is incompatible with run-te/run-tm, or if the dielectric tensor is not positive-definite.
  • Default to vendor cc instead of gcc, so that C and Fortran compilers are in sync. (We default to the vendor f77 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 compute-energy-in-object-list for non-orthogonal lattices.
  • Bug fix in combine-band-functions 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 floating-point sensitivity bug in mpb-data that could cause a crash on the Alpha; thanks to Dominique Caron for the bug report and debugging information.

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 --with-inv-symmetry; 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 eigensolver-block-size 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 (zero-frequency) 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 k-point (instead of once per band per k-point as for ordinary band functions).
  • compute-field-energy function now also returns the fraction of the energy in the various field components. Thanks to Karl Koch for the suggestion.
  • The filename-prefix 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 first-brillouin-zone function to transform an arbitrary k-point into an equivalent point in the first Brillouin zone. Thanks to Payam Rabiei for the suggestion.
  • In mpb-data, 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 spherical-quadrature 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.

MPB 0.12

July 9, 2000

  • Added fix-*field-phase functions to allow a deterministic phase in the output fields, thanks to a suggestion by Doug Allan.
  • Added group-velocity calculation functions (display-group-velocities, etcetera).
  • Added -e x,y,z option to mpb-data 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 for epsilon 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.
  • compute-field-energy now returns the total, unnormalized energy in the corresponding field; combined with compute-energy-in-objects, this makes it easy to do some perturbation theory and related calculations.
  • Eigensolver improvements. Periodic reorthogonalization and renormalization to combat some numerical problems. New line-minimization code, included with permission from MINPACK-2 by Jorge J. Moré.
  • Fixed breaking of 90-degree rotational symmetry-breaking by the mesh in 2d; thanks to Jim West and Doug Allan of Corning for the bug report. (In general, some symmetry-breaking 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.

MPB 0.11

February 12, 2000

  • configure script can now detect and link ATLAS 3.0 accelerated BLAS.
  • Added band-range-data output variable.
  • Running mpb-data multiple times on the same file now replaces the results of the previous run, instead of appending -new2, -new3, etcetera.
  • Fixed bug in run-even/run-odd that could seriously slow or even prevent eigensolver convergence. Thanks to Payam Rabiei for the bug report.
  • Fixed compilation --without-hdf5, or when HDF5 is not found. Thanks to Rajesh Rengarajan for the bug report.

MPB 0.10

January 28, 2000

  • Added mpb-data utility for post-processing data (e.g. for unskewing non-orthogonal lattices). See the new data-analysis tutorial or man mpb-data for more information.
  • Added new data-analysis tutorial to the manual, describing how to analyze and visualize the results of two sample calculations.
  • Added support for a new material type, dielectric-anisotropic, so that you can specify arbitrary real/symmetric dielectric tensors.
  • Added new output-at-kpoint function to make it easier to output fields only at a single k-point in a band-structure 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 k-point vectors to facilitate post-processing.
  • 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 less-standard INCLUDES to pass -I flags to the configure script (for header files in non-standard locations).
  • Added diamond.ctl example file for a 3d diamond (fcc) lattice of spheres.
  • Added (brief) mpb man page.
  • Fixed z-parity output and run-even/odd functions for 2d grids.
  • Fixed bug in output-dpwr-in-objects. Thanks to Mihai Ibanescu for the bug report.
  • Compilation fixes. We need to set SHELL in the Makefile for make on some systems. Also added rule to insure ctl-io.h is created before main.c is compiled. Thanks to Christoph Becher for the bug reports.

MPB 0.9.1

January 7, 2000

  • Fixed eigensolver bug where special handling of Gamma (k=0) point could screw up convergence for subsequent k-points, causing incorrect results.
  • Fixed behavior of filename-prefix input variable; thanks to Karl Koch for the bug report

MPB 0.9

January 2, 2000

  • Added run-even and run-odd 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 new z-parity output variable.
  • Added epsilon-input-file 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 breaking make install on some systems. Thanks to Ron Chase for the bug report.

MPB 0.8.1

November 22, 1999

  • Added output-hfield-x, output-dfield-y, etcetera, functions for outputting only specific field components.
  • Sped up HDF5 field output routines.
  • Added output-copies variable to set the number of periods output by the band output functions.

MPB 0.8

November 19, 1999

  • Initial public release.
Personal tools