Dev#23
Merged
Merged
Conversation
… knob
Satellite stripping (strip_from_satellite) removed 1/N of the recomputed
baryon excess per substep, so the fraction stripped over a snapshot is
1-(1-1/N)^N -- set by the substep count N rather than elapsed time or any
physical timescale. Under adaptive/dynamical-time stepping N varies, so this
becomes a redshift-dependent artifact in principle.
Add an opt-in physical alternative behind PhysicalStrippingOn (default 0):
0 = legacy geometric (excess/effective_steps), reproduced bit-for-bit so
the calibrated baseline is unchanged
1 = physical timescale: strip excess*(dt/t_strip) per substep, telescoping
to 1-exp(-dT/t_dyn) over a snapshot -- N-invariant in the limit and
tied to the host dynamical time (Rvir/Vvir), passed in from the
substep loop. Per-substep fraction capped at 1 for the dt>=t_strip
(high-z) limit.
Add SubstepResolution (double, default 1.0): a runtime multiplier on the
adaptive substep count (scales both the STEPS floor and MAX_STEPS cap)
without resizing the compile-time SFR arrays, enabling convergence /
N-invariance sweeps from the parameter file instead of recompiling.
Add plotting/compare_stripping.py: overlays N runs on the four
stripping-sensitive diagnostics (satellite quenched fraction, gas fraction,
HI mass function by type, baryon fraction vs Mvir) for A/B testing.
Tests: existing 16 stripping calls keep asserting legacy behaviour on the
default path; add 3 physical-path tests (exponential law, N-invariance,
high-z cap). All 18 suites pass.
Verified in production runs: at matched N legacy and physical are
indistinguishable (<0.2% in every satellite statistic); the only residual
is a small global resolution effect shared by both schemes.
…n knob
Legacy satellite stripping removed 1/N of the recomputed baryon excess per
substep, so the fraction stripped over a snapshot is 1-(1-1/N)^N -- set by the
substep count N rather than elapsed time. Under dynamical-time stepping N
varies, making this an in-principle redshift/cadence artifact.
Add two opt-in physical alternatives behind PhysicalStrippingOn (default 0):
0 = legacy geometric (excess/effective_steps), reproduced bit-for-bit so
the calibrated baseline is unchanged.
1 = physical timescale, per substep: strip excess*(dt/t_dyn) each substep,
telescoping to 1-exp(-dT/t_dyn); N-invariant in the limit (~1% residual).
2 = physical timescale, analytic: strip exactly 1-exp(-dT/t_dyn) once per
snapshot, OUTSIDE the substep loop -- zero substep-count dependence and
cadence-invariant (the Option-1 "evaluate once" + Option-2 "exponential"
hybrid). t_dyn is the central's Rvir/Vvir.
Add SubstepResolution (double, default 1.0): runtime multiplier on the adaptive
substep count (scales STEPS floor and MAX_STEPS cap) for convergence /
N-invariance sweeps without recompiling or resizing the SFR arrays.
Add plotting/compare_stripping.py to overlay N runs on the stripping-sensitive
diagnostics (satellite quenched fraction, gas fraction, HI mass function by
type, baryon fraction vs Mvir).
Tests: 16 existing stripping calls keep asserting legacy behaviour on the
default path; add 5 physical-path tests (exponential law, N-invariance, high-z
cap, analytic exactness, substep-count independence). All 18 suites pass.
Verified in Millennium runs: modes 0/1/2 agree to <0.2% in every satellite
statistic at production resolution (stripping saturates); the analytic mode is
the substep-independent, cadence-invariant formulation to prefer going forward.
… with dynamical time to strip outside of the substep loop. Removing stripping being dependant on time-steps.
…d really just be kept as 1.0, just the dynamical time.
# Conflicts: # .gitignore # src/core_build_model.c
Baselines now reflect the merged physics (physical stripping schemes + HII ionization). 5444 datasets per config, determinism re-verified (two consecutive runs bit-identical).
- Remove committed Sphinx build output (docs/_build) and Doxygen XML (_doxygen/) from version control; they are build artifacts. - Ignore docs/_build/ and _doxygen/ going forward. - Drop editor/tool-local ignore entries that are better kept in .git/info/exclude.
- Fix root 'make tests' target (was pointing at a nonexistent launcher script); add 'make regression' for the byte-for-byte baseline sweep. - Write the missing docs/developer/REGRESSION_BASELINE.md referenced by the regression scripts; document the two verification systems and the bit-identical policy. - Move root-level developer notes into docs/developer/. - Reconcile dataset-count claims (5380 -> 5444) in README, docs, and CONTRIBUTING. - Point run_sage_batch.sh at the shipped .par files. - Ignore generated docs (docs/_build/, _doxygen/).
The committed reference.sha256 predated the physical-stripping rework and the HII-ionization merge; physics on this branch legitimately differs. Regenerated with mpirun -n 1; verify now passes.
The PROCESS_LHVT_STYLE conditional was never defined, guarded by an #error, and called functions that do not exist anywhere in the tree. Remove the dead blocks in sage.c, the orig_index field they used, and core_tree_utils.c/.h whose three public functions were only referenced from those blocks. Verified: all 5444 baseline datasets bit-identical.
USE_SAGE_IN_MCMC_MODE was not defined by any build configuration and diverged from the fiducial output (2 of 82 HDF5 fields, altered merger bookkeeping). Keep only the standard paths. Verified: all 5444 baseline datasets bit-identical.
- Delete #if 0 blocks (disabled short-Type struct variants, ctrees column self-test, stale #undef). - Delete commented-out debug fprintf lines across the tree readers, HDF5 writer, and allocator. - Replace an embedded code-generation prompt transcript in buffered_io.c with the module's standing header description. - Gate the per-task forest-assignment [LOG] stderr prints behind VERBOSE like every other info message. Verified: all 5444 baseline datasets bit-identical.
- GALAXY::CentralMvir was never read or written; the output writers compute their own CentralMvir from the FOF host halo. - GALAXY::TotalSatelliteBaryons was accumulated every snapshot but never read by physics or either output format; drop the field, the accumulation loop, and its self-contained unit test. - params::CGMrecipeSAGEOn was declared but never parsed or referenced. Verified: all 5444 baseline datasets bit-identical; all 18 unit test suites pass.
- Make file-local functions static (do_AGN_heating_cgm, the two BR06 disk-structure helpers) and drop their public declarations. - Replace magic 60.0/30.0 literals in starformation_ffb with the existing FIRE_V_CRIT_KMS / KD11_METAL_HALO_MASS constants (same double values, bit-identical). - Use MAX_STRING_LEN-1 instead of a stale literal 511 in the parameter parser bounds. - Drop the inert -DGNU_SOURCE flag (nothing tests GNU_SOURCE; glibc keys on _GNU_SOURCE and gnu99 already enables GNU extensions). - Fix inaccurate comments: NDEBUG claims in macros.h, evolve_galaxies docstring, ConcentrationOn mode list, stale GD14 cross-references. - Normalize the remaining tab-indented lines in the physics files. Verified: all 5444 baseline datasets bit-identical; all 18 unit test suites pass.
starformation_ffb computed the identical (1+z)^alpha * (V/Vcrit)^beta scaling twice (reheating and ejection). Hoist it into a single precomputed value, mirroring the main star-formation path. Same expression, operands, and evaluation order. Verified: all 5444 baseline datasets bit-identical.
generate_galaxy_indices repeated the same multi-line hint five times, differing only in reader filename and long-stale hardcoded line numbers. Replace with one generic message (plus the consistent-trees ascii variant). Error-path stderr text only. Verified: all 5444 baseline datasets bit-identical.
The 'CGM when CGMrecipeOn and Regime==0, else HotGas' if/else ladder was repeated at eight deposit sites across star formation, mergers, and reincorporation. Replace with two static inline helpers (add_metals_to_hot_reservoir / add_gas_to_hot_reservoir) in model_misc.h. The += operations are unchanged. Verified: all 5444 baseline datasets bit-identical; all 18 unit test suites pass.
The 'tracks molecular gas' test appeared as a six-term OR chain in some places and as (sf != 0 && sf != 2) in others; the BR06 family test appeared as (== 1 || == 3). Replace all with sf_prescription_tracks_h2() / sf_prescription_is_br06() inline predicates. SFprescription is now validated to [0, 7] at parameter read (previously an out-of-range value silently formed zero stars), which makes the two historical forms provably equivalent. Verified: all 5444 baseline datasets bit-identical; all 18 unit test suites pass.
- New GitHub Actions workflow: serial build + unit tests, MPI compile check, and a non-blocking mini-Millennium smoke run (trees cached; external download must not gate merges). Dependencies come from a single conda prefix via micromamba because the Makefile's CI path derives HDF5_DIR from gsl-config --prefix. - Fix the two gcc -Wformat-truncation/-Wstringop-truncation warnings that would trip -Werror on CI (memcpy of the fixed-size tag buffer; precision-capped %s). Warning-clean under both clang and gcc-16. Verified: all 5444 baseline datasets bit-identical.
Profiling (macOS sample) shows the run dominated by libm transcendentals, with calculate_molecular_fraction_radial_integration the hottest SAGE function (25 bins per galaxy per substep): - BR06 (default): the stellar scale height (pow/log10) depends only on the disk scale length; compute once per galaxy instead of per bin via a _from_hstar variant. The per-bin arithmetic is unchanged. - KMT09/K13: the metallicity, chi, and the log numerator depend only on disk metallicity; hoisted before the loop (identical values). - KD12: reuse the hoisted disk metallicity. Also update tests/README.md counts to the actual 559 tests across 18 suites and refresh the per-suite table. Verified: all 5444 baseline datasets bit-identical; all 18 unit test suites pass. Median runtime 11.53s -> 11.10s (~4%) on mini-Millennium.
The sage_binary writer writes GALAXY_OUTPUT structs to disk verbatim. The struct has 8 bytes of alignment padding per record (before g_max and after r_heat) which was never initialised, so uninitialised heap bytes leaked into the output files. Every field value was unaffected, but the files were non-reproducible at the byte level: the benchmark checksum test failed intermittently (~1 in 4 runs), on this branch and on the pre-cleanup base commit alike. memset the output block before filling it and regenerate the benchmark reference (only padding bytes changed; verified 0 science-byte diffs against the previous output and 10/10 clean verify runs). HDF5 output is written per-field and was never affected. Verified: all 5444 HDF5 baseline datasets bit-identical; all 18 unit test suites pass.
Add the three satellite-stripping parameters (PhysicalStrippingOn, SubstepResolution, StrippingTimescaleFactor) at their compiled defaults so the reference config lists every registered parameter, and normalise LastSnapshotNr spelling to match the code (parsing is case-insensitive, so no behaviour change either way). Verified: all 5444 baseline datasets bit-identical.
A syntactically valid but fake ORCID is worse than none; the CFF schema does not require it. Re-add when a real iD is available.
The smoke lanes never passed: the mini-Millennium tarball holds trees/... relative paths and was being unpacked at the workspace root instead of inside input/millennium (where first_run.sh unpacks it), so sage found no tree files. The MPI build check failed because conda-forge packages the MPI compiler wrappers separately -- installing openmpi alone provides no mpicc; add openmpi-mpicc and the compilers toolchain. Also add workflow_dispatch so the workflow can be exercised from a branch, bump the tree cache key past any treeless cache entries, and verify the unpack produced trees_063.0 before caching.
The tarball holds flat files (trees_063.*, millennium.a_list) with no directory prefix -- first_run.sh untars it from inside input/millennium/trees/, not one level up. The unpack-verification guard added in the previous commit caught this.
Caught by AddressSanitizer on the (short-lived) sanitized CI lane: the SaveFullSFH Description/Units attributes passed string literals to CREATE_STRING_ATTRIBUTE with a declared size of MAX_STRING_LEN, so H5Awrite read ~1 KiB from a ~100-byte literal -- an out-of-bounds read whose trailing bytes (whatever rodata followed the literal) were written into the output file. The per-field Description/Units buffers were correctly sized but uninitialized past the snprintf'd text, leaking stack bytes into the file the same way the struct-padding bug did for binary output. Literals are now staged through a zeroed MAX_STRING_LEN buffer and the field-metadata arrays are zeroed before filling. Attribute datatypes and file layout are unchanged; every byte written is now defined. All 5444 datasets verified bit-identical.
The mini-Millennium smoke run and its sanitized variant depended on downloading simulation trees from an external mirror inside CI, which is fragile and pulls data through GitHub runners on every push. Both lanes served their purpose (the sanitized run caught the HDF5 string-attribute overflow fixed in the previous commit); model runs are a local gate, and sanitized runs remain available locally via 'make MEM-CHECK=yes'. CI keeps the hermetic jobs: serial build + unit tests and the MPI compile check.
…unit tests Round 3 of the pre-release cleanup: units audit and code-unit macros, per-paper SF prescription functions, model_misc split by physics domain, startup option validation, clamp counters, the approved HI ionisation bookkeeping fix (baselines re-captured), H2 chemistry unit tests, the HDF5 string-attribute overflow fix, and CI repairs (MPI wrapper package, tree unpack path) with the download-dependent smoke lanes removed.
New model_ram_pressure.c implements the Gunn & Gott (1972) criterion for stripping satellite ColdGas: gas at disk radius r is removed where P_ram = eps * rho_host(R_orb) * v_sat^2 exceeds the restoring force 2*pi*G*Sigma_disk(r)*Sigma_gas(r). With exponential gas and stellar disks sharing DiskScaleRadius the stripping radius is analytic and the stripped fraction is (1 + r_strip/rs)*exp(-r_strip/rs). The driver runs once per snapshot from evolve_galaxies(), outside the substep loop, with the same 1-exp(-dT/t_strip) cadence and t_strip as the PhysicalStrippingOn == 2 hot-gas scheme; the two channels are otherwise independent (halo starvation vs ISM removal). Ambient density comes from the same profiles the cooling recipes assume: cgm_density_at_radius() (now external) for CGM-regime hosts, isothermal M_hot/(4 pi Rvir r^2) for hot-regime hosts. Stripped gas and metals route to the central via the regime-aware add_gas_to_hot_reservoir(). New optional parameters (registered and range-checked at startup): RamPressureStrippingOn (0/1, default 0) and RamPressureEpsilon (default 1.0, order-unity infall-geometry prefactor). With the default RamPressureStrippingOn = 0 the millennium and microuchuu baselines verify bit-identical.
Covers ram_pressure_stripped_fraction (bounds over 12 decades of P_ram, guard values, monotonicity, f_strip = 1 at/above the central restoring force, ~0 in the deep-potential limit, exact (1+x)exp(-x) at x = 2) and ram_pressure_strip_satellite (mass and metal conservation between the satellite and the host reservoir, metallicity of the stripped gas, regime-aware routing to CGMgas, the 1-exp(-dt/t_strip) cadence factor, no-ambient-medium / gas-free / zero-disk-radius guards, the Vvir fallback for degenerate velocities, and weaker stripping at larger orbital radius in the isothermal host profile).
Adds a ram_pressure_strip_satellite() section to docs/physics/infall.md (criterion, ambient profiles, analytic stripped fraction, once-per- snapshot cadence, the fixed-rs surface-density caveat, and the calibration-check warning for toggled-on runs) and registers RamPressureStrippingOn / RamPressureEpsilon in the parameter tables.
RamPressureStrippingOn = 0 and RamPressureEpsilon = 1.0, the built-in defaults, so the millennium_all baseline stays bit-identical (verified: 5444 datasets).
plotting/rps_validation.py compares matched microUchuu runs (toggle off / epsilon sweep) against the validation-plan observables: quenched Type-1 satellite fraction vs host mass (Wetzel+12 SDSS band), median satellite HI-to-stellar ratio vs host mass (Brown+17 trend), the global HI mass function against the ALFALFA Schechter fit (Jones+18) as the calibration-safety check, and median HI retention vs time since infall split by host mass -- the cumulative-exposure signature of the mechanism. Missing sweep runs are skipped so the script works on any subset.
Orphans dominate cluster cores but were skipped, capping the HI deficiency the channel can produce exactly where observations show the strongest signal. Include them with a frozen-orbit approximation: the position stays frozen at the snapshot the subhalo was lost (the true orbit only decays further in, so rho_host and hence the stripping are underestimated -- the conservative side), while the stale stored velocity is replaced by the host Vvir, since its offset from the CURRENT central velocity is not a meaningful orbital speed. Adds a unit test pinning the orphan path to the equivalent Type 1 satellite at v = Vvir. Default-off output unchanged (millennium verify bit-identical).
Type 2 orphans merge or disrupt within a snapshot of subhalo loss in the default configuration (zero survive to any output), so the frozen-orbit path only matters for configurations with long-lived orphans.
do_AGN_heating_cgm clamps coolingGas against the pre-accretion CGMgas, then subtracts the Bondi AGNaccreted from the reservoir -- so the returned coolingGas can exceed the remaining CGM by up to AGNaccreted and the caller overdraws it. Latent with the default uniform profile (the overlap of a reservoir-limited cooling clamp and nonzero accretion stays below the 1e-12 assert tolerance); CGMDensityProfile = 1 (NFW) trips the XASSERT abort immediately at startup. Millennium regression remains bit-identical; full unit suite passes.
The precipitation branch drains the CGM at dM/dt = f_precip * M_CGM /
t_ff once tcool/tff < PRECIP_THRESHOLD, so the only stable end state is
an empty reservoir: 99% of CGM-regime centrals at logMvir 11.3-12 sit
below the threshold (median tcool/tff ~ 0.2) and dump their whole CGM
into the disk, where H2-thermostated SF cannot consume it. This is the
dominant source of the factor 2-90 HIMF excess over ALFALFA and the
+0.2-0.45 dex offset from the xGASS gas fractions (the excess is 80%
centrals). The CGM density profile does not matter: NFW and beta runs
move the census by ~1% because the sigmoid saturates regardless.
PrecipRegulationOn = 1 condenses only the gas above the Voit
equilibrium: t_cool scales as 1/M_CGM at fixed profile and temperature,
so M_eq = M_CGM * (tcool/tff)/threshold and
dM/dt = f_precip * (M_CGM - M_eq) / t_ff
relaxes toward tcool/tff = threshold and shuts off, converting the
free-fall dump of the stored reservoir into a pass-through limited by
the rate at which infall and reheated gas push the CGM back over M_eq.
Default 0 is byte-identical (millennium verify passes). Unit tests
cover the exact throttle factor and the self-quenching relaxation.
BHs currently bootstrap only through mergers and disk instabilities, so merger-poor field halos arrive at the Mshock crossing with undermassive BHs (median logM_BH 6.7 vs 7.4 for normal 1e12 centrals) and no radio-mode maintenance -- they keep cooling at 14-38 Msun/yr and build the logMHI > 10.5 tail of the HI mass function (x50-90 over ALFALFA, unmoved by every wind/cooling/reincorporation lever tested). BlackHoleSeedOn = 1 gives every new galaxy a BlackHoleSeedMass seed (default 1e-5 = 10^5 Msun/h) so the Bondi radio-mode loop can engage in quiet halos. EXPERIMENTAL placeholder for the dedicated BH-seeding work; default 0 leaves output bit-identical (millennium verify passes).
This reverts commit 3a4e302.
…efault Flip RamPressureStrippingOn and PrecipRegulationOn to on by default after the microUchuu validation this branch built: - RamPressureStrippingOn = 1: Gunn & Gott (1972) ISM stripping of satellites reproduces the satellite HI-deficiency-vs-host-mass trend and lifts the cluster-scale quenched fraction toward Wetzel+12. - PrecipRegulationOn = 1: precipitation relaxes toward the tcool/tff equilibrium instead of draining the CGM at the free-fall rate; the physically correct Voit behaviour (macroscopically small at z=0 since little CGM sits in the marginal window, but correct). millennium_all.par updated to pin both on explicitly; the gitignored vanilla configs are pinned off locally to preserve classic-SAGE runs. This intentionally changes the default output -- baselines regenerated in the following commit.
Intentional baseline update to bless the on-by-default ram-pressure stripping + precipitation regulation as the canonical output. Recaptured with the serial build (make USE-MPI=): - tests/baseline/millennium (5444 datasets), verify passes - tests/baseline/millennium_all (5444 datasets), verify passes - tests/benchmark/reference.sha256 (64 binary files), verify passes The local microuchuu baseline (gitignored) was recaptured too. These replace the pre-toggle byte-identical references; classic-SAGE output remains reproducible via the vanilla configs.
Ram-pressure ISM stripping + precipitation regulation (now default-on)
The default flip (c33be43) updated the RamPressureStrippingOn struct comment and the read-parameter default comments but left several places still describing the old off-by-default state: - core_allvars.h PrecipRegulationOn struct comment (said 0=DEFAULT) - model_ram_pressure.c header (said default 0) - docs/parameters.md RamPressureStrippingOn default column (0 -> 1) and a missing PrecipRegulationOn row (now added) - docs/physics/infall.md two RamPressureStrippingOn default mentions Comment/doc only -- no output change.
docs: fix stale default-off comments for RPS / precipitation regulation
Bring README, CHANGELOG, and the RTD physics docs up to date with the now-default-on channels merged in #20/#21: - CHANGELOG.md: new top entry covering ram-pressure ISM stripping, the self-regulating precipitation law, the AGN/CGM overdraw bugfix, and the intentional default flip with regenerated baselines. - README.md: RamPressureStrippingOn + PrecipRegulationOn added to the 'what is new' table, a satellite ram-pressure stripping subsection, and the precipitation-regulation row in the CGM table. - docs/physics/cooling_and_heating.md: precipitation self-regulation (M_eq / free-fall shut-off), the PrecipRegulationOn switch, and the post-accretion coolingGas re-cap. parameters.md and infall.md were already updated in #20/#21. RTD build verified locally (doxygen + sphinx, all pages render clean).
…bsteps, disruption split Backfill feature-level changelog coverage that was missing: - Physical satellite stripping (PhysicalStrippingOn schemes 0/1/2 + StrippingTimescaleFactor): the substep-count-dependent stock scheme vs the physical 1-exp(-dT/t_strip) schemes, with the analytic once-per-snapshot scheme 2 as default (the timescale/cadence RPS reuses). - Adaptive substeps (SubstepResolution): substep count scales with deltaT/t_dyn so high-z snapshots are resolved; runtime multiplier for convergence testing. - Satellite-disruption mass split (DynamicDisruptionSplit + alpha/cref): fixed / mass-ratio / concentration-weighted ICS-vs-central partition. RTD build verified. HI ionisation was already covered by the HI bookkeeping section.
…bsteps Backfill the physics-options coverage that was missing from the README, matching the changelog: - Satellite hot-gas stripping (PhysicalStrippingOn schemes 0/1/2 + StrippingTimescaleFactor): the starvation channel complementary to the ISM ram-pressure stripping, default scheme 2 (analytic, substep-invariant). - Adaptive time integration (SubstepResolution): substep count scales with deltaT/t_dyn; runtime multiplier for convergence testing. DynamicDisruptionSplit was already in the Mergers/ICS table; the two new default-on channels (RPS, precip regulation) were added earlier.
…knob The substep sweep showed SubstepResolution is the single largest lever on the massive-end SMF (coarse steps over-cool), and its runtime cost makes it a real decision rather than a hidden default. Expose it explicitly in millennium.par under a dedicated 'Numerical time resolution' section with a prominent calibration-locked / runtime-vs-convergence comment, and document it (plus the previously-undocumented PhysicalStrippingOn row and StrippingTimescaleFactor) in docs/parameters.md with a convergence note: ~0.1 dex SMF shift from 1.0 to converged (within observational scatter), runtime ~linear in substep count, effect inherited from classic SAGE cooling (not the new physics). Adds the default value only -> output byte-identical (millennium verify passes).
Hardcode the calibrated behaviour and drop the parameters (each fixed at its former default, so output is byte-identical): - PhysicalStrippingOn -> always the analytic once-per-snapshot scheme (former default 2); scheme 0/1 branches and the in-loop strip call removed. strip_from_satellite loses its effective_steps argument. - StrippingTimescaleFactor -> t_strip = t_dyn(host) (former 1.0). - PrecipRegulationOn -> precipitation is always self-regulating (former default 1); the free-fall-drain branch removed. - HIIonizationOn -> the HI ionisation cut is always applied (former 1). - SigmaHIcrit -> fixed SIGMA_HI_CRIT constant = 0.5 Msun/pc^2. - CGMAGNOn -> CGM-regime AGN heating fires whenever AGNrecipeOn > 0 (former 1); the condition drops the CGMAGNOn term. Struct fields, REG() entries, defaults, and range-checks removed; millennium_all.par tags dropped. Unit tests updated to the single behaviour (stripping scheme tests collapsed to the analytic case; HI partition and precipitation tests rewritten toggle-free). Byte-identical verified: millennium, millennium_all, microuchuu (all datasets), and the MPI benchmark (64 files). 20/20 unit suites pass.
Update the parameter reference, physics docs, README, and changelog for the now-fixed behaviour: drop the PhysicalStrippingOn / StrippingTimescale Factor / PrecipRegulationOn / CGMAGNOn parameter rows; rewrite the satellite hot-gas stripping section for the single analytic scheme; state precipitation self-regulation and the HI ionisation cut (Sigma_HI_crit = 0.5) as always-on; add a CHANGELOG entry. RTD build verified.
# Conflicts: # .gitignore
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.