feat(deep_causality_homology): Added homology crate - #755
Merged
Conversation
…mology crate Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
…crate Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
…eep_causality_homology
Homology was split across two crates. `deep_causality_topology` owned the concepts —
`ChainComplex`, `HomologyField`, `Gf2Chain` — and delegated every computation to
`deep_causality_linear`, which carried conversion helpers whose docstrings named
topology's boundary operators as their reason to exist. Neither crate owned the seam.
The cost landed with the QCL work. A quantum error-correcting code is a chain complex
that is not a space: `H_X` and `H_Z` are parity-check matrices with no cells, no metric
and no Hodge star. Reaching 419 lines of chain-complex machinery meant depending on
27,317 lines of geometry.
`deep_causality_homology` 0.1.0 now holds that layer: the `ChainComplex` trait over
boundary matrices alone, `HomologyField`, and `Gf2Chain<W>`. Its whole dependency set is
`deep_causality_linear` and `deep_causality_num`.
The trait splits in two. Six homology items move; the five geometry items — `CellType`,
`CellIter`, `Metric`, `cells`, `uniform_lattice_layout` — stay in topology on a new
`CellularComplex: ChainComplex` supertrait. Not named `CellComplex`, because that is a
published struct in the same crate root and Rust puts types and traits in one namespace.
`boundary_matrix` keeps `Cow<'_, CsrMatrix<i8>>`. Incidence numbers lie in {-1, 0, 1} by
construction, so `i8` is an invariant of the boundary operator rather than a storage
choice, and the trait needs no coefficient parameter.
`Gf2Chain` keeps `(degree, len)` as the identity of its chain group. `C_k = F_2^{n_k}` is
fixed by the cell count, and every operation the type offers belongs to that group rather
than to a complex. The two halves of that condition were checked in two places and raised
two different error types; `same_group` now checks both and raises one.
Four defects, found while moving:
1. Basis orientation. `kernel_basis_gf2` allocates `zeros(cols, free.len())` and writes
basis vector k down column k. Four docstrings said rows, and `from_row` is a
contiguous word-slice copy that cannot read a column, so a caller following them got
a vector of the wrong length. Adds `PackedGf2Vector::from_column` and `Gf2Chain::
from_column`; corrects the docstrings.
2. Degenerate grades returned an empty matrix. All three implementors now give the
shape the dimension implies — d_0 is (0, n_0) and d_{max+1} is (n_max, 0) — so
cols(d_k) == rows(d_{k+1}) holds at both ends. `betti_number_over` survived the old
behaviour on `saturating_sub`; a kernel basis would not.
3. `d . d = 0` was stated nowhere. It is now a documented law on the trait and an
assertion in the conformance harness at every grade of every implementor, with
coefficients widened past `i8` so a wrapped intermediate cannot read as zero.
4. `test_boundary_matrix_squared_is_zero` asserted cols(d_2) == rows(d_1) — that is n_2
against n_0 — which held on square_torus(2) only because that torus has four 0-cells
and four 2-cells. Having checked a shape it returned without forming the product its
name promises. Both repaired.
Formalization. `linear.gf2.betti_from_ranks` proved the Betti identity under a hypothesis
nothing supplied: `range d_{k+1} <= ker d_k`, which is the chain condition. Every Betti
number the workspace computed rested on an assumption written down once, as an argument
to a theorem. `lean/DeepCausalityFormal/Homology/ChainCondition.lean` discharges it:
`homology.chain.dd_zero_implies_range_le_ker` turns it into a matrix identity a test can
check, and `homology.chain.betti_from_dd_zero` restates the Betti identity over that.
Zero `sorry`. Both ids carry Rust witnesses that compute the two sides by separate
routines. See `deep_causality_homology/LEAN_HOMOLOGY.md`.
Test references are external. `openspec/notes/homology/reference/reference.py` builds ten
spaces in Python, checks its own Betti numbers against Hatcher, *Algebraic Topology*, and
imports nothing from this workspace. The crate's fixtures are an independent construction
of the same spaces checked against the same published values. RP^2 and the Klein bottle
are included because every complex the workspace shipped before this change is orientable
and torsion-free, so Q and F_2 agreed everywhere and the coefficient field was never
discriminated; beta_1(RP^2) is 0 over Q and 1 over F_2.
Verification: bazel test //... 1227/1227; clippy clean; cargo mutants over the new crate
50 mutants, 40 caught, 10 unviable, 0 survivors.
MIGRATION
`deep_causality_topology` re-exports `ChainComplex`, `HomologyField` and `Gf2Chain`, so
`use deep_causality_topology::ChainComplex` keeps working and no dependent manifest
changes. All four dependents pin version = "0.7" and pick this up untouched.
Two call sites need a one-line import change, and no re-export avoids it: a method belongs
to one trait, and calling it needs that trait in scope. Code that calls `cells()` or
`uniform_lattice_layout()` through a `ChainComplex` import must import `CellularComplex`
instead, which has `ChainComplex` as a supertrait:
-use deep_causality_topology::ChainComplex;
+use deep_causality_topology::CellularComplex;
Both are in `deep_causality_cfd` and are updated here. Callers of `num_cells`,
`max_dim`, `boundary_matrix`, `coboundary_matrix` and the Betti methods are unaffected.
`ChainComplex::betti_number_over` returns `Result<usize, HomologyError>` in place of
`TopologyError`. Only implementors that override it are affected; every implementor in
this workspace uses the provided body.
`widen_to_dense_i64`, previously private to topology, is now
`deep_causality_linear::csr_i8_to_dense_i64`.
Versions: deep_causality_homology 0.1.0 (new), deep_causality_linear 0.1.1 -> 0.1.2,
deep_causality_topology 0.7.3 -> 0.7.4.
Implements openspec/changes/extract-homology-crate.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
`add-linear-algebra-crate` moves to `openspec/changes/archive/`, and its capabilities
become the established baseline under `openspec/specs/`: 72 requirements added across 13
new capabilities, 1 modified.
New: linear-crate-identity (4), linear-matrix-representations (6), linear-vector (6),
linear-solve (5), linear-dense-algorithms (7), linear-f2-algebra (5),
linear-integer-algebra (5), linear-scalar-contract (4), linear-tower-integration (5),
linear-hkt-composition (4), linear-consumer-migration (7),
linear-test-first-development (9), num-finite-field (5).
Modified: `neumann-poisson` already existed in `openspec/specs/`, and the change carried
it as MODIFIED rather than ADDED, so it merged instead of clobbering. The preconditioned
CG requirement is re-homed from `deep_causality_sparse` to `deep_causality_linear` —
same signature, same convergence behaviour, same iteration counts — and gains a scenario
pinning that both import paths resolve to one function while the retired crate still
re-exports it.
Eight tasks were open. Both groups are closed on evidence, and the evidence is recorded
in the archived tasks.md rather than implied by the checkmark.
Group 7, publishing, verified against the crates.io API rather than from memory:
deep_causality_linear 0.1.1 (2 versions), deep_causality_sparse 0.2.5 (13),
deep_causality_tensor 0.5.3 (24), deep_causality_topology 0.7.3 (16),
deep_causality_physics 0.8.2 (17). Zero yanked versions across all five, which closes
7.5 directly. For 7.4, published deep_causality_topology 0.7.3 carries
deep_causality_linear at req = ^0.1, resolving to the published 0.1.1; `cargo publish`
builds the tarball with path dependencies stripped, so its acceptance is the
resolve-and-compile check that task asks for.
Group 8 said "file separately", so the action was to file, not to fix. The three
findings are now in `openspec/notes/linear/FOLLOW-UPS.md`, re-verified against the tree
first — all three still hold, and one had drifted:
- deep_causality_physics `invert_3x3` (gr_utils.rs:114) and `inverse_spatial_metric`
(adm_state.rs:126) reject a singular spatial metric at 1e-14 and 1e-12, a factor of
100 apart, and the second compares in f64 — lossy for Float106, the extended
carrier the ADM state can be instantiated at.
- `CausalMultiField::inverse` documents "Uses matrix inverse for each cell" and maps
the multivector reversion inverse instead.
- `CausalTensor::matmul` is over-bounded on PartialOrd. The cited path had moved to
ops/tensor_product/mod.rs:13, and the same over-bound also sits on the trait method
at api/mod.rs:35 — which the original note missed. Matrix multiplication needs no
ordering, and the bound excludes the complex numbers, quaternions and octonions,
all of which are Ring.
None of the three is fixed here, and the note says so.
Archiving invalidated three references to the pre-archive path, all repointed:
- `extract-homology-crate/proposal.md` justified its empty Modified Capabilities with
"the linear crate's specs remain under the unarchived add-linear-algebra-crate
change", which stopped being true. The reason that survives is that the promoted
linear-* specs describe the matrix and elimination layer the homology crate
consumes, not the chain complex it defines over them.
- `openspec/notes/linear/prototype/README.md` and
`openspec/notes/linear/deep-causality-linear.md` pointed at the live change
directory.
Also ignores `__pycache__/`, produced by running the reference oracles under
`openspec/notes/*/reference/`.
`openspec validate extract-homology-crate` passes. No source file is touched.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
`extract-homology-crate` moves to `openspec/changes/archive/`, and its three capabilities become the baseline under `openspec/specs/`: 15 requirements, 39 scenarios, no collisions with existing specs and nothing modified. homology-chain-complex 5 requirements, 13 scenarios homology-gf2-chain 5 requirements, 14 scenarios topology-cell-complex-seam 5 requirements, 12 scenarios One requirement was wrong and is corrected before promotion. `topology-cell-complex-seam` carried "Existing consumers compile without an edit", with a scenario asserting dependents build with "no edit to any `use` statement". Implementation disproved that: a re-export carries a name and cannot carry a method to a trait that no longer owns it, so code calling `cells` or `uniform_lattice_layout` through a `ChainComplex` import must name `CellularComplex`. The proposal and tasks were corrected when the build caught it; the spec was not, and archiving would have installed a guarantee the code does not make. It now states the real boundary. A consumer of the homology half — `num_cells`, `max_dim`, `boundary_matrix`, `coboundary_matrix`, `betti_number`, `betti_number_over` — compiles with no edit at all. A consumer of a geometry method needs the owning trait named, and that one import line is sufficient: no call site, signature or bound changes. The measured split is recorded with it: eighteen `num_cells` sites and every `deep_causality_physics` file untouched, two sites in `deep_causality_cfd` needing the import. Task 10.7 closes with what happened rather than a bare checkmark: the commit message was prepared and handed over, and committed by the author as 5d591e9. Repoints every reference to the notes trees that moved to `openspec/notes/archive/`. Twenty files carried a dead path — four in the specs and changes archived here and in 8154e98, twelve in `deep_causality_linear` and `deep_causality_homology` sources and tests, and the rest in `deep_causality_sparse` and the mathematics examples. All are comments, docstrings and prose; `cargo check` over the three affected crates is clean. `openspec/notes/linear/HKT-LAW-FINDINGS.md` is the one exception to the mapping — it went to `openspec/notes/unified_math/`, not the archive, and is repointed there. No source behaviour changes. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
`extract-homology-crate` moves to `openspec/changes/archive/`, and its three capabilities become the baseline under `openspec/specs/`: 15 requirements, 39 scenarios, no collisions with existing specs and nothing modified. homology-chain-complex 5 requirements, 13 scenarios homology-gf2-chain 5 requirements, 14 scenarios topology-cell-complex-seam 5 requirements, 12 scenarios One requirement was wrong and is corrected before promotion. `topology-cell-complex-seam` carried "Existing consumers compile without an edit", with a scenario asserting dependents build with "no edit to any `use` statement". Implementation disproved that: a re-export carries a name and cannot carry a method to a trait that no longer owns it, so code calling `cells` or `uniform_lattice_layout` through a `ChainComplex` import must name `CellularComplex`. The proposal and tasks were corrected when the build caught it; the spec was not, and archiving would have installed a guarantee the code does not make. It now states the real boundary. A consumer of the homology half — `num_cells`, `max_dim`, `boundary_matrix`, `coboundary_matrix`, `betti_number`, `betti_number_over` — compiles with no edit at all. A consumer of a geometry method needs the owning trait named, and that one import line is sufficient: no call site, signature or bound changes. The measured split is recorded with it: eighteen `num_cells` sites and every `deep_causality_physics` file untouched, two sites in `deep_causality_cfd` needing the import. Task 10.7 closes with what happened rather than a bare checkmark: the commit message was prepared and handed over, and committed by the author as 5d591e9. Repoints every reference to the notes trees that moved to `openspec/notes/archive/`. Twenty files carried a dead path — four in the specs and changes archived here and in 8154e98, twelve in `deep_causality_linear` and `deep_causality_homology` sources and tests, and the rest in `deep_causality_sparse` and the mathematics examples. All are comments, docstrings and prose; `cargo check` over the three affected crates is clean. `openspec/notes/linear/HKT-LAW-FINDINGS.md` is the one exception to the mapping — it went to `openspec/notes/unified_math/`, not the archive, and is repointed there. No source behaviour changes. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #755 +/- ##
========================================
Coverage 97.22% 97.23%
========================================
Files 1319 1322 +3
Lines 77875 78121 +246
========================================
+ Hits 75716 75959 +243
- Misses 2159 2162 +3 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
4 issues found across 199 files
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="lean/BUILD.bazel">
<violation number="1" location="lean/BUILD.bazel:44">
P2: The new Homology namespace is registered for the Bazel gate but the root aggregator lean/DeepCausalityFormal.lean does not import DeepCausalityFormal.Homology.ChainCondition, while it imports every other namespace's files. CI (formalization.yml) runs only `lake build`, which builds the DeepCausalityFormal root module and its imports, so this proof is never type-checked by the actual CI gate. Add `import DeepCausalityFormal.Homology.ChainCondition` to the root aggregator.</violation>
</file>
<file name="openspec/specs/num-finite-field/spec.md">
<violation number="1" location="openspec/specs/num-finite-field/spec.md:49">
P2: The motivating claim that the three multivector sites 'sit under a `Field` bound' today is stale: in the current tree all three (`commutator_geometric`, `to_coefficients`, `commutator_geometric_impl`) are already bounded on `DivisibleByIntegers`, with doc comments that restate this spec. The cited line numbers (mod.rs:163, conversions.rs:139, ops_product_impl.rs:316) also no longer match where the `T::one() + T::one()` terms live (181, 145, 326). Update the present-tense problem statement and citations to the current code so the spec doesn't misrepresent what still needs migrating.</violation>
</file>
<file name="openspec/specs/linear-consumer-migration/spec.md">
<violation number="1" location="openspec/specs/linear-consumer-migration/spec.md:102">
P3: The file counts are stale: the literal currently appears in 129 files, with 44 under openspec/changes/archive/, not 203 and 34. Refresh the numbers so the requirement and its archive-touchstone rationale match the tree.</violation>
</file>
<file name="deep_causality_topology/src/types/manifold/mod.rs">
<violation number="1" location="deep_causality_topology/src/types/manifold/mod.rs:48">
P2: This bound rejects chain-only `K` types, but the module, README, and HKT documentation still promise `K: ChainComplex`. Update those public descriptions to require `CellularComplex` and explain the narrower API.</violation>
</file>
Reply with feedback, questions, or to request a fix.
Re-trigger cubic
… and 42 review findings Code review raised 35 findings against the homology extraction and the specs it promoted. Each was verified against the tree before any edit; 34 confirmed, 8 confirmed with a wrong detail, none refuted. Every reviewer claim held up in substance. BREAKING: deep_causality_topology 0.7.4 -> 0.8.0 The published 0.7.3 `ChainComplex` carried six geometry items and returned `TopologyError`. The extraction removed the geometry items, moved them to `CellularComplex`, and changed the error type to `HomologyError` — then shipped that as a patch. `git show 5d591e9^` has the old trait. A re-export carries a name, not a trait's contents, so this breaks: - an external `impl ChainComplex for MyType` defining `CellType`, `CellIter`, `Metric`, `cells` or `uniform_lattice_layout` now fails with E0437, - a `K: ChainComplex` bound naming `K::CellType` or calling `k.cells(..)` no longer resolves, - `betti_number_over` returns a different error type. All three re-export shims said the old import path kept working. They now carry a `# This is a breaking change` section naming what moved and what a caller does about it. Four in-workspace manifests pinned topology at "0.7" and would have failed to resolve; physics, cfd, algorithms and discovery are repinned to "0.8". CODE - `Neighborhood`, `FaceAdjacent` and `CofaceAdjacent` go back to a `ChainComplex` bound. An automated pass widened them to `CellularComplex` during the split, but the bodies touch only `max_dim` and the boundary and coboundary matrices, so the wider bound excluded abstract complexes for nothing. - `LatticeComplex::boundary_matrix` returned `num_cells(D)` rows at every grade above the top, not only at `D + 1`, so `cols(∂_{D+1}) == rows(∂_{D+2})` was false. Grades past `D + 1` now return `(0, 0)`. The conformance harness did not reach above `max_dim + 1`; it does now, and the fix was temporarily reverted to confirm the widened harness fails without it. - `ChainComplex::betti_number_over` stepped the grade with `saturating_add(1)`, so at `k == usize::MAX` it re-read `∂_MAX` and subtracted its rank twice. It steps with `checked_add` and takes a next-boundary rank of zero at the ceiling. - `reference_spaces()` covered nine of the oracle's ten spaces. `T³` is added, gated `#[cfg(not(miri))]`: at 702 simplices against the next-largest 96 it takes 50 seconds to build under Miri against 2 milliseconds native, so `make miri` skips it the way other crates skip their slow suites. FORMALIZATION `lean/DeepCausalityFormal.lean` imported ten namespaces and not `Homology`, so `Homology/ChainCondition.lean` was reachable from no built module. CI runs `lake build`, which builds the root and its imports, so the proof was never type-checked by the gate that was supposed to check it. The Bazel target passed and the CI gate was hollow. The header's two module lists also omitted `Homology`, `Rational` and `Quantum` while importing all three; all are described now, and the lists are verified against the imports. `ChainCondition.lean` is warning-clean. Chasing the linter shrank the proof: `Matrix.mulVecLin_mul` is already a `simp` lemma, so the `have` bridging the composite was doing nothing and `simp [h]` closes the goal alone. `Homology` is renamed `HomologyGroup`, since inside namespace `DeepCausalityFormal.Homology.ChainCondition` the old name produced a duplicated path. `reference.py` returned `(0, 0)` for the top degenerate grade where `d_{max+1}` has one row per maximal cell, contradicting the file's own emitted shape table and the contract the Rust crate implements. Pasting those shapes into the conformance suite would have pinned the wrong top-grade shape. `betti()` also selected 𝔽₂ for any field name that was not exactly `Q`, so a typo silently computed over the wrong field; unknown names are now rejected. TEST FIXTURES ARE HIDDEN FROM THE DOCUMENTED SURFACE `utils_tests` must be `pub` because Bazel cannot reach the `tests` tree from a test target. It does not have to be documented. Eight crates rendered it on docs.rs as public API: algebra, haft, linear, num_complex, physics, topology, ultragraph and tensor. All are `#[doc(hidden)]` now, joining homology, which already was. Two of them needed more than the attribute. A public function in `cell_splitting.rs` linked into the module, and rustdoc renders a hidden module when public documentation points into it. Haft's crate-level doc advertised `utils_tests` in a hand-written module tour, which no attribute affects. Verified by rendering: `utils_tests` appears on none of the nine crate indexes, and `axis_mask` is still documented at its real public path. SPECS Eleven promoted specs and three archived-change documents described a pre-refactor tree. Corrections include: a cited source path in a retired crate; a baseline witness credited with `Monad` and `Adjunction` that it deliberately does not implement; a "two marker impls away" ring gap that is already closed, where following the spec would have produced an E0119 duplicate impl; hardware-tied benchmark ratios standing as normative pass/fail gates with no bench target to reproduce them; a justification citing an `svd()` call and a `1e-5` threshold that exist nowhere in the tree; scope arithmetic that summed to 14 and 17 where the text said 13 and 16; a conjugate-linearity scenario whose assertions any symmetric form satisfies; a `TBD` placeholder; and stale file counts, recounted. Two claims about this change's own specs were also wrong and are fixed in both the archived and promoted copies: `deep_causality_algebra` was listed in a dependency set that holds two crates, and a scenario required the row-oriented basis path to be absent or to report a mismatch when it is public and silently returns a different length. `linear-vector` claimed length mismatches fail with a typed error rather than panicking. The checked `add`, `sub` and `dot` do return errors; `core::ops::Add` fixes `type Output = Self` and has no error channel, so it asserts. Both paths are stated. Four relative links were broken by the notes move and are repointed. VERIFICATION bazel test //... — 1227/1227. clippy --workspace --all-targets — 0. fmt --check — clean. cargo doc --workspace — 0 errors. openspec validate --specs — 172/172. bazel test //lean:Homology --nocache_test_results — passes, 0 sorry. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
Contributor
There was a problem hiding this comment.
All reported issues were addressed across 58 files (changes from recent commits).
Reply with feedback, questions, or to request a fix.
Re-trigger cubic
…our stale references
A second review pass over the homology work raised five findings. All five were verified against
the tree before any edit and all five held.
THE LATTICE FIXTURE IS NOW ONE IMPLEMENTATION
`torus_3` carried its own facet-generation and validation pipeline because `lattice_quotient` was
written for two dimensions. `lattice_quotient` is const-generic over `D` now, and all five lattice
fixtures are one call each:
lattice_quotient("torus_2", [3, 3], |[x, y]| [x % 3, y % 3])
lattice_quotient("torus_3", [3, 3, 3], |[x, y, z]| [x % 3, y % 3, z % 3])
`[usize; D]` rather than slices, because the call sites decide it. Arrays let the gluing closures
destructure, so coordinates keep their names and the Klein-bottle and Mobius closures stay the
one-liners they were. A `&[usize] -> Vec<usize>` signature would have made them
`|p| vec![p[0] % 3, p[1] % 3]`: unnamed coordinates, a heap allocation per lattice point, and an
arity mistake that panics at run time instead of failing to compile.
The axis-order enumeration replaces a `filter(|[a, b, c]| a != b && b != c && a != c)` that only
worked at three axes. The degeneracy assertion is kept at every dimension rather than adopting
reference.py's `if len(s) == dim + 1`, which drops a folded simplex silently; the Rust fixture is
now stricter than the Python oracle on that point.
This also removes a latent bug. The old `torus_3` used one constant for both the grid size and the
axis count, so a 4x4x4 torus would have generated 24 index triples over `0..4` and panicked
incrementing a corner. `sizes: [usize; D]` separates the two roles.
The refactor is behaviour-preserving, and that is measured rather than inferred. Every non-zero
boundary-matrix entry, with its sign and position, at every grade of all ten fixtures, is identical
before and after: the fixture module was swapped back to its previous revision, fingerprinted, and
diffed. The vertex labels did not move either, `torus_3` included, because the mixed-radix fold
reduces to the base the old code hardcoded.
FOUR STALE REFERENCES
- `chain_complex.rs` carried the `usize::MAX` boundary-case comment twice, pasted verbatim. One copy
remains.
- `qcl-gaps.md` cited the three topology re-export shims at lines 22, 15 and 15. The `pub use`
statements are at 38, 25 and 27; the old numbers landed a reader in doc-comment prose. The file
was swept for recurrences and has none.
- `linear-f2-algebra` pinned the same re-export to `homology_field/mod.rs:15`, also a doc comment.
Now `:25`.
- `linear-dense-algorithms` said both determinant sites are fed "matrices the diagonal cannot
pivot". That is true of one. The Cayley-Menger matrix has a zero diagonal end to end: the buffer
starts at zero, `one` is written only into `1..matrix_dim`, and every remaining diagonal entry is
a vertex's squared distance to itself. The Gram matrix's diagonal holds squared edge-vector
norms, positive whenever the vertices are distinct and zero only when one repeats, so it is
usable but poorly scaled. The replaced `gaussian_determinant` tested it against an absolute
floor, which read a uniformly small simplex as degenerate for being small; the shared
`determinant` scales its floor by the matrix's own magnitude. The two sites are now described
separately.
Signed-off-by: Marvin Hansen <marvin.hansen@gmail.com>
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.
Describe your changes
feat(deep_causality_homology): Added homology crate
The homology extracts abstract chains so the quantum crate
can use them without pulling in the entire topology crate.
The topology crate was refactored to use the chain traits
from the homology crates.
Issue ticket number and link
Code checklist before requesting a review
For details on make, please see BUILD.md
Note: The CI runs all of the above and fixing things before they hit CI speeds
up the review and merge process. Thank you.
Summary by cubic
Adds
deep_causality_homology, a new crate holding the chain-complex layer (chain complex trait, homology field, mod-2 chain) so the quantum crate can use homology without depending on the topology crate's geometry.Extracts the homology half of
deep_causality_topology::ChainComplexinto the new crate; geometry items (CellType,CellIter,Metric,cells,uniform_lattice_layout) move to a newCellularComplextrait withChainComplexas a supertrait. Topology re-exportsChainComplex,HomologyField, andGf2Chainso name imports keep compiling, but this is a breaking topology release (0.7.3 → 0.8.0): externalimpl ChainComplexblocks that defined geometry items no longer compile, andbetti_number_overreturnsHomologyError.Migration
cells()oruniform_lattice_layout()must importCellularComplex;deep_causality_cfdis updated.widen_to_dense_i64is nowdeep_causality_linear::csr_i8_to_dense_i64;deep_causality_linearbumps 0.1.1→0.1.2.Bug fixes
(0,0)so the shape contract holds, andbetti_number_overno longer double-counts atusize::MAX.from_column, degenerate grades return shape-implied matrices,∂∘∂=0is documented and asserted, and a boundary-squared test that never formed its product is repaired.ChainCondition.lean) is now wired into the build — it was previously never type-checked.utils_testsfixtures are#[doc(hidden)]across eight crates so they no longer render as public API.lattice_quotientshared by all five lattice spaces, removing a latenttorus_3bug where one constant served as both grid size and axis count.Written for commit 2fe7100. Summary will update on new commits.