Changelog
2.1.0 (2023-10-03)
Port to TRIQS 3.2.
2.0.0 (2022-06-27)
Major changes and new features
Complete port to TRIQS 3.1 and Python 3.
Implementation of the Stochastic Optimization with Consistent Constraints (SOCC) proposed by Goulko et al in Phys. Rev. B 95, 014102 (2017). It includes three major pieces of functionality.
The Consistent Constraints update in the Markov chain used to accumulate particular solutions;
The Consistent Constraints protocol for constructing final solutions out of particular solutions;
The solution quality assessment technique implemented in a new Python module
som.spectral_stats
.
For consistency with MaxEnt and other stochastic continuation methods, the objective function of the optimization problem has been changed to the “goodness of fit” \(\chi^2\)-functional.
Adoption of the \(\chi^2\)-functional has made it possible to support user-supplied covariance matrices of input data as an alternative to simple estimated error bars (credits to @snirgaz for proposing this feature).
A new family of integral kernels for symmetric fermionic Green’s functions has been introduced. The corresponding observable is called
FermionGfSymm
.The
BosonAutoCorr
kernels have been changed to more closely reproduce results of theBosonCorr
kernels for the same input data. Both kernel families are defined on the whole energy axis and expect the same spectrum normalization constants now (before one had to divide the constants by 2 forBosonAutoCorr
).Projection of an observable onto a real frequency mesh can now be performed using binning (enabled by default). In this mode the projected observable is integrated over bins centered around points of the mesh.
Further MPI parallelization.
Massively reworked online documentation.
Python API changes
Following a convention change for TRIQS applications, the Python package of SOM has been renamed from
pytriqs.applications.analytical_continuation.som
to a laconicsom
.Functionality of the
run()
method ofSomCore
has been split among a few new methods,accumulate()
– accumulate particular solutions;adjust_f()
– adjust the number of global updatesF
;compute_final_solution()
– construct the final solution using the standard SOM protocol;compute_final_solution_cc()
– construct the final solution using Goulko’s SOCC protocol.
One may still call the deprecated
run()
, which is equivalent to callingaccumulate()
+compute_final_solution()
.In recent versions of TRIQS it became impossible to use the
<<
syntax to fill GF containers from user-defined Python objects. Furthermore, high-frequency tail data was separated from the GF containers. As a result, that syntax had to be abandoned in favor of a few free functions.fill_refreq()
– fill a real-frequency observable from a computed SOM solution;compute_tail()
– compute high-frequency tail coefficients from a computed SOM solution;reconstruct()
– reconstruct input from a computed SOM solution.
It is now possible to resume accumulation of particular solutions by calling
SomCore.accumulate()
multiple times, and to discard all accumulated solutions by callingSomCore.clear()
.A handful of new properties and accessor methods have been added to
SomCore
.The
rectangle
andconfiguration
C++ objects are now exposed as Python classesRectangle
andConfiguration
.Configuration
objects can be saved to/loaded from HDF5 archives.Updated signature of
som.count_good_solutions()
to take bothgood_chi_abs
andgood_chi_rel
(thresholds on \(\chi\) and \(\chi/\chi_\mathrm{min}\) for a solution to be considered good).A new utility function
som.estimate_boson_corr_spectrum_norms()
has been added. Given a correlator of boson-like operators \(\chi\) defined on any supported mesh, it returns a list of spectrum normalization constants \(\mathcal{N} = \pi \chi(i\Omega = 0)\).
Build system and developer tools
Minimum required CMake version has been bumped to 3.12.4.
Structure of the project has been adjusted to follow conventions established by the app4triqs application template.
A
Dockerfile
has been added.Files
som.modulefile
andsomvars.sh
are generated and installed as part of the build process.New benchmarks:
all_kernels
,binning
,consistent_constraints
,bosonautocorr
andfermiongfsymm
.The
chi
benchmark has been removed as it depended on the privatetriqs_ctseg
code.Support for C++ static analysis tools
clang-tidy
andcppcheck
has been added.A CMake option has been added to link
libsom
and unit tests to Clang sanitizers (AddressSanitizer
andUndefinedBehaviorSanitizer
).C++/Python coding style is enforced with
clang-format
andflake8
.
1.2 (2020-03-15)
Improvements and small fixes in documentation.
Added a new histogram post-processing function,
count_good_solutions()
.Fixed a bug in
update_glue_shift
elementary update.New benchmark
all_kernels
and fixes in thechi
benchmark.Added Travis CI config for continuous testing and documentation deployment.
Minor code improvements.
1.1 (2017-04-23)
Massive extension of documentation.
adjust_f
mode is disabled by default.Fixed a critical bug in
back_transform()
.
1.0 (2017-03-19)
First official release.
Supports 12 integral kernels:
4 observable kinds (
FermionGf
,BosonCorr
,BosonAutoCorr
, andZeroTemp
);3 input meshes (
imtime
,imfreq
,legendre
).