Release 24.08

24.08

• autodiff is now used with the templated reaction networks (#1614)

  • some autodiff clean-ups and derivative fixes (#1604, #1612,
    #1613, #1616, #1619, #1633)

• we can now output warnings from GPUs if you compile with
  USE_GPU_PRINTF=TRUE (#1629, #1635)

• documentation improvements (#1570, #1628)

• a new jacobian unit (jac_cell) test was added that compares the
  numerical and analytic Jacobians (#1618)

• support for Strang + NSE has been removed. NSE only works with
  SDC now (#1549, #1621)

• the network CNO_He_burn was added for explosive H/He burning
  (#1622)

• code clean-ups (#1582, #1602, #1609, #1623, #1624, #1625, #1626,
  #1627, #1631, #1639)

• test_nse_net now also tests the NSE EOS interface (#1621)

• the self-consistent NSE + SDC update has been synced with the
  tabular NSE implementation (#1569, #1607, #1617)

• test_jac was not correctly evaluating the numerical Jacobian
  (#1615)

• the fast_atan function is now more accurate (#1611)

• test_ase was renamed test_nse_net and the old test_nse was
  removed (#1610)

• the old test_screening unit test was removed (#1608)

• the RKC integrator now supports NSE bailout (#1544)

• a second temperature check for tabular NSE was added -- above this
  temperature we don't consider composition (#1547)

• a SDC+NSE unit test was added (#1548)

• a fast log and fast pow approximation was added (#1591)

• the primordial_chem network now uses the fast math routines (#1605)

• fix potential Inf in constexpr linear algebra (#1603)

Release 24.07

24.07

• added an autodiff library and converted all of the screening
  functions to use autodiff for the thermodynamic derivatives
  (#1581, #1588, #1593, #1596, #1597, #1600)

• some testing infrastructure fixes (#1598, #1589)

• documentation improvements (#1594)

• added approximate math functions for exp and atan (#1583, #1586)

• fix return code for PrimordialChem unit test (#1590)

• NSE optimizations (including chabrier1998 screening) #1585

• remove "using namespace amrex" from most headers (#1584)

• NSE table can work with other network now (#1576, #1577, #1580)

• the subch_full and subch_approx networks were removed -- these
  are replaced by subch_simple and subch_base (#1578)

• retry tolerances now default to use the same values as the first
  attempt, unless they are explicitly set in an inputs file (#1573)

Release 24.06

24.06

• added the ability to access the runtime parameters via a struct.
  This will eventually be used to remove the dependency on globals
  (#1433, #1575)

• simplified the integrators by extracting common infrastructure
  into setup and cleanup functions (#1515, #1546)

• lots of documentation improvements including sectioning (#1559)
  integrators (#1560, #1561, #1562, #1563, #1566, #1567, #1568),
  runtime parameters (#1557), and link checking (#1552)

• CUDA no longer requires VODE + generalize some AMREX_USE_CUDA
  to AMREX_USE_GPU (#1564)

• self-consistent NSE now accepted the temperature threshold as a
  runtime parameter (#1558)

• general code cleanups (#1537, #1551, #1553, #1554)

• unit tests no longer append .cxx to output (#1309)

• added an eos_rh_t EOS type (#1539)

Release 24.05

24.05

• Runtime parameters can now be type bool (#1536)

• more clang-tidy and compiler warning cleaning (#1527 #1530,
  #1532, #1533)

• Remove recursion in quicksort to avoid CUDA stack limits (#1531)

• Update the pynucastro networks to cache derived rate partition
  functions (#1529)

Release 24.04

24.04

• A new test_screening_templated unit test was added -- this
  works with any of the templated networks. (#1525)

• A lot of small code clean-ups from clang-tidy (#1516, #1518, #1519, #1520,
  #1522)

• The NSE solver was optimized (#1503, #1504, #1506, #1507, #1508)

• The integrator code was synced up between implementations, fixing
  a bug in the RKC retry tolerances (#1513)

• A reinterpret_cast in rhs.H was removed (#1435)

Release 24.03

24.03

• pivoting in the linear algebra routines can now be disabled
  (#1454)

• the scaling of the energy derivatives in the Jacobian when
  running with integrator.scale_system=1 has been fixed (#1479)

• added a new linear algebra unit test (#1493)

• when building with HIP we disable forced inlining (#1490)

• improved the energy update with NSE and remove unused terms
  (#1483, #1484, #1485)

• remove using namespace amrex from most headers (#1465, #1474)

• updated the pynucastro networks to pynucastro 2.2.0 (#1470)

• fixed an error code check in the VODE integrator (#1472)

• added a zone-by-zone retry capability to the burner (#969)

Release 24.02

24.02

• Lots of general code cleaning from coverity and clang-tidy
  (#1450, #1452, #1453, #1460, #1459, #1457, #1458)

• Fixed a bug in the VODE pivoting when a cached Jacobian is used
  (#1456)

• Added reverse rates to CNO_extras (#1445)

• Sync networks up with pynucastro to get constexpr mion/bion
  (#1437)

• NSE+SDC improvements (#1431)

• Start of moving the runtime parameters from globals to structs
  (#1422)

Release 24.01

24.01

• The quantum corrections for the Chabrier screening are
  now optional (#1428)

• We've replaced std::pow() with amrex::Math::powi for integer
  powers for better GPU performance (#1432)

• in_nse now works with an eos_t for tabular NSE (#1424)

• There are a new set of interfaces for inverting the EOS when
  we are in NSE (with the tabular NSE) that consistently find
  T and the composition (#1405, #1430)

• The NSE table now uses finer spacing (#1427)

• The SDC+NSE update for tabular NSE is now based on a 2nd-order
  Runge-Kutta method (#1415)

• An additional check on molar fractions was added to
  self-consistent NSE to determine if we are in NSE (#1417)

• The NSE table interface was changed (#1404, #1418)

• A script that checks if a network is compatible with
  self-consistent NSE was added (#1416)

• constant T evolution was fixed (#1408)

• An unsafe reinterpret_cast was removed from linear algebra (#1412)

• Methods for computing T derivatives of an NSE table quantity were
  added (#1407)

Microphysics 23.12

23.12

• The SDC+NSE update now includes plasma neutrino losses (#1357,
  #1400)

• The default tabular NSE interpolation is now cubic (#1399)

• Self-consistent NSE now requires chabrier1998 or null screening
  (#1398)

• A new network, subch_base, was added that further simplifies
  subch_simple (#1393)

• A slightly larger network for Urca was added (#1365)

• A new NSE table was added. This is generated via pynucastro and
  there is a python script that can be used to regenerate it (#1350)

• A bug was fixed in the neutrino cooling that was introduced in an
  optimization last release (#1380)

Microphysics 23.11

23.11

• The sneut5 neutrino cooling term was cleaned up (#1371, #1372,
  #1373, #1374, #1375, #1377, #1378, #1379)

• The number of predictor-corrector iterations for the SDC+NSE algorithm
  is now a runtime parameter (#1370)

• The Urca network now includes a more accurate rate for neutron decay
  and electon-capture onto a proton. (#1359)

• The He-C-Fe-group network now includes the positron parts of the
  weak reaction rates (#1360)

• A check was added to ensure that the helm_table.dat is valid on
  reading (#1355)