Skip to content

Latest commit

 

History

History
388 lines (251 loc) · 12.3 KB

CHANGES.md

File metadata and controls

388 lines (251 loc) · 12.3 KB
Raw

20.05

  • We now have an option for using sparse storage for aprox13 in C++ (#307)

  • iso7 and aprox13 are now available as a C++ network (#303, 305)

  • species names are available as an enum in network_properties.H (#304)

  • The screening on O16+O16 in iso7 was fixed (#302)

  • The VODE integrator is now available in C++ (#299)

20.04

  • The wion network property was removed (#294)

  • There are new unit tests for the screening and aprox rates modules (both C++ and Fortran interfaces).

  • The screening routines were ported to C++ (#290) and the screenz routine was removed in favor of screen5 (#293)

  • a new method, is_input_valid, was added to all EOSes (both C++ and Fortran interfaces) that can be used to query whether an EOS supports a particular input mode (e.g. eos_input_rp). (#291)

  • The aprox rates used with iso7, aprox13, aprox19, and aprox21 have been converted to C++ (#288)

  • We've rewritten the VODE integrator to remove all "go to" statements (#275, 276, 278, 280, 281, 282, 283, 284, 285, 286, 287)

  • We removed the ability to have nspec_evolve < nspec. This feature was not widely used and greatly complicated the code paths.(#279)

20.03

  • The nuclei information for both Fortran and C++ is now automatically generated from a network inputs file at compile time. As part of this change, 1/A is precomputed and stored as a constant (#253, 258).

  • The license for StarKiller Microphyscs was made explicit and a license.txt file was added (#267)

  • A framework for pure C++ EOSes has been created and a pure C++ unit test, test_eos_C, is available to test these. (#246) The following EOSes have been ported to C++: ztwd (#268), multigamma (#265), polytrope (#264), gamma_law (#263), helmholtz (#262), gamma_law_general (#246), rad_power_law (#269), breakout (#270)

  • The GPackage.mak files that were a remnant of the old BoxLib F90 build system have been removed. They were not maintained. (#212).

  • All of the interface files have been collected together in the interfaces/ dir. (#240)

  • The network C++ headers have been renamed network_properties.H and the nuclei information (aion and zion) have been added. (#244)

20.02

  • Added a C++ header file, actual_network.H, that defines the network size. This is the start of making making the microphysics routines available in C++.

  • regenerated the pynucastro networks with the latest weak rate formulations from pynucastro.

20.01

  • The burn_t type no longer includes ydot or jac -- this allows us to optimize the memory access on GPUs (#220)

  • The radiation pressure contribution to the Helmholtz EOS has had a dampener applied to it that makes it approximately zero for low densities where the radiation pressure would lead to unphysical situations like a superluminal sound speed. (#235)

  • The original VODE integrator was removed and the Fortran 90 version VODE90 was renamed to VODE. (#221)

  • The test_react unit tests no longer require a composition inputs file (xin*). They now create the composition profile at runtime. (#211)

19.12

  • Simplified SDC integration now uses the same retry strategy as the default (non-SDC) integration. (#215)

  • VODE90 can now participate in the retry strategy that was previously available to the VODE integrator, where it can switch to the BS integrator if loosening the tolerances does not allow the burn to complete. (#201)

  • The parameter ode_max_steps was made consistent in VODE and VODE90; in some places it was being ignored. (#214)

  • The helmholtz EOS was restructured, splitting the different components into different functions and optimizing the memory accesses. (#200)

  • The derivatives with respect to mass fraction (dpdX, dedX, dhdX) were removed from eos_t and are now available through a new type, eos_xderivs_t and the composition_derivatives() routine. (#207)

  • A bug in the screening of the C12+C12 and O16+O16 rates in iso7 was fixed.

  • The test_eos unit test now outputs all of the variables in the eos_t type.

19.11

  • VODE90 now works with the simplified SDC time step algorithms, and the preprocessor option for this SDC was changed to SIMPLIFIED_SDC (#194)

  • rprox now works on GPUs

19.10

  • The iso7 network was ported to GPUs (#172)

  • VODE90 now better agrees with VODE (#192)

  • When building with multiple integrators, the contents of the rpar modules could clash. This has been fixedc (#136)

  • A module for making the "Nonaka plot" tracking the evolution of a quantity during the burn was added, and is enabled with USE_NONAKA_PLOT=TRUE

19.08

  • A new network, subch2, was added that combines aprox13 and the subch networks. (#184)

19.05

  • The aprox21 network was missing the analytic Jacobian term for the derivative of He4 with respect to Ni56. This is fixed. (#175)

  • The numerical Jacobian module, used by the BS and VBDF integrators had some wrong scalings. These have now been fixed (#179, #180)

19.01

  • the docs are now automatically build from the sphinx source using travis on github.

18.12

  • simplify conductivity interface to match the eos interface by moving the conductivity into the eos type

18.11

  • new python bindings have been added

  • the documentation has been switched to Sphinx and is now hosted online.

  • a bug was fixed in the stellarcollapse EOS where the energy offset was being applied incorrectly.

18.10

  • test_eos and test_react now both work on GPUs (using the AMReX gpu branch)

  • the intermediate blending of the weak and strong screening regimes was wrong, and has been fixed. We've also synced some parameters up to agree with those in MESA and Kepler. (#149, 150)

  • eos_input_is_constant is now set to true for the helmholtz EOS. This mean that the EOS inputs will not be modified after the EOS call. This is good for conserving energy in a hydro code, but the tradeoff is a small (to root finding tolerance) inconsistency in the thermodynamic state.

18.09

  • The Helmholtz parameters ttol and dtol (controlling the error for the Newton iteration when in a mode other than eos_input_rt) are now runtime parameters in the extern namelist as eos_ttol and eos_dtol.

18.08

  • the unit tests (test_react, test_sdc, and test_eos) have been ported from the Fortran to C++ build system in AMReX. This will allow us to test the GPU framework in AMReX.

18.07

  • added CUDA support to the VODE90 integrator, the helmeos, and the networks aprox13, aprox19, aprox21, ignition_simple, C-burn-simple, URCA-simple.

  • Ported the unit test frameworks to FBoxLib

18.05

  • lots of documentation updates

  • some fixes to the numerical Jacobian involving X vs. Y

  • a new subCh network for He burning was added.

  • implemented the new c12(a,g)o16 nuclear reaction rate and its corresponding inverse from the work of Deboer et al. 2017 (https://journals.aps.org/rmp/abstract/10.1103/RevModPhys.89.035007). To use the new rate, user must set use_c12ag_deboer17 to true. This rate is only useable in the aprox13, aprox19, aprox21, and iso7 reaction rate networks. Closes issue #44.

  • a routine util/cj_detonation was added to compute the Chapman-Jouguet detonation velocity for any of the networks

  • the burn retry strategy now sticks with the current integrator and uses looser tolerances before switching to a different integrator.

18.04

17.11

  • a new option to boost the reaction rates has been added to the integrators (PR #64)

  • we now disable some composition derivatives in the EOS by default, for performance and memory reasons. They can be reenabled by defining the preprocessor variable EXTRA_THERMO (PR #59)

17.10

  • the compositional derivatives are no longer available by default from the EOS. To get these, set the preprocessor variable EXTRA_THERMO. This change was done for performance reasons.

  • the aprox19 and aprox21 networks no longer use a numerical Jacobian by default, as this was found to result in some bad numerical issues in VODE (PR #49)

  • the maximum temperature for reactions, MAX_TEMP, is now an adjustable input parameter rather than being hardcoded at 1.d11.

  • the Helmholtz EOS table is now read by the IO processor and broadcast to other processors (PR #53)

  • the VODE integrator now does some additional checks on the state to ensure consistency (PR #47)

17.09

  • a new rety mechanism was implemented that allows a different integrator to be used if the primary integrator fails

  • the electron Ni56 electron capture rates and energy losses were updated from Mazurek (1973) to LMP (2000). Thanks to Carl Fields for this contribution. Pull request #40

17.08

  • fix to aprox21 from Aron Michel (HITS) that fills in missing reactions

  • updated the helmholtz EOS to use the latest table from Frank Timmes (in particular, this is now denser, with 2x points in T and rho dimensions). If you copied the old table, you need to make sure you are using the new table now.

  • add stellar conductivities from Frank Timmes

17.06

  • a new Fortran 90 port of VODE has been added

  • the unit tests now require AMReX instead of BoxLib to build

17.01

  • we've removed the option to integrate molar fractions and instead the ODE system always operates on mass fractions (the networks return derivatives of molar fractions and they are automatically converted).

16.12

  • a new unit test, test_sdc, was created to test the SDC interface to the networks

  • we now rely on the network module to provide aion_inv (1/aion)

  • the VODE integrator now supports SDC integration

16.09

  • num_rate_groups is now a property of the individual networks

  • a new integration method, Rosenbrock, was added to the BS option (set ode_method)

  • the number of RHS and Jac evaluations is now passed out of the burner through the burn_t type for diagnostic and load-balancing use

  • support for spectral deferred correction coupling of the burner and hydro was added to the BS integrator

16.08

  • Microphysics/eos/ has been renamed Microphysics/EOS/ to better conform to the conventions used in Castro and Maestro

  • the User's Guide has been extensively updated

  • OpenMP and OpenACC directives have been added to the unit tests

  • the BS integrator's type, bs_t, has now contains a burn_t internally, simplifying the conversion from bs_t to call the actual_rhs/jac

  • the rates() component of burn_t was removed. We no longer rely on rate caching

  • we now store the simulation time at the start of the burn as t0 in the rpar storage to use as an offset to the integration time

  • the species derivatives (dh/dX and de/dX) and enthalpy were removed from the burn_t

  • a new option to integrate of X instead of Y was added (integrate_molar_fraction = F)

  • integration of networks with nspec_evolve < nspec were fixed to now apply the algrebic constraint relating mass fractions through a new update_unevolved_species() function

  • the electron capture rate on Ni56 used by aprox19 and aprox21 was fixed

  • the BS integrator can now use the initial timestep estimation algorithm that VODE uses 9use_timestep_estimator = T)

  • a centered difference numerical Jacobian option was added

16.07

  • we now use MICROPHYSICS_HOME instead of MICROPHYSICS_DIR as the environment variable to point to the Microphysics/ directory.

  • there are now two standalone unit tests, test_react and test_eos that don't need Maestro or Castro to compile.

  • new burn mode that limits numerically unstable burning.

  • UsersGuide/ was renamed to Docs/ to be consistent with the other BoxLib codes

  • the energy equation now uses an offset to help with the BS ODE integration convergence

  • the runtime parameter small_x now is owned by the network