initial attempts at adding a smooth canon for maximum

Ipopt interface prototype

adding more smooth canonicalizers

* adds oracles and bounds class to ipopt interface

* adds some settings and solver lists changes for IPOPT

* adds nlp solver option and can call ipopt

* adds more experiments for integrating ipopt as a solver interface

* passing the problem through the inversion

* add some more extra changes

* adding nlmatrixstuffing

---------

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

* adding many tests, new smoothcanon for min, and improvements to ipopt_nlpif

* fixing last two tests

* add another example, qcp

* adding example for acopf

* add control of a car example done

---------

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

* update solution statuses thanks to odow

* removes unusued solver information

---------

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

* getting rocket landing example to work

* add changes to the jacobian computation

---------

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

* adding many more example of non-convex functions

* making lots of progress on understanding good canonicalizations

---------

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

Co-authored-by: William Zijie Zhang <william@gridmatic.com>

* Rename ESR/HSR to linearizable_convex/linearizable_concave

Spell out opaque acronyms for clarity per PR review feedback:
is_atom_esr → is_atom_linearizable_convex,
is_atom_hsr → is_atom_linearizable_concave,
is_esr → is_linearizable_convex, is_hsr → is_linearizable_concave,
is_smooth → is_linearizable.

Docstrings clarify that "linearizable convex" means the expression is
convex after linearizing all smooth subexpressions.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Adopt three-way atom classification: smooth, nonsmooth-convex, nonsmooth-concave

Replace the two-axis is_atom_linearizable_convex/concave overrides across all
atoms with a single method from the paper's three categories: is_atom_smooth,
is_atom_nonsmooth_convex, or is_atom_nonsmooth_concave. The base class derives
the old linearizable methods for backward compatibility.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Inline atom-level linearizable checks into expression-level composition rules

Remove the intermediate is_atom_linearizable_convex/concave methods and
_has_dnlp_classification helper, which were only used in the two
expression-level composition rules in atom.py. The three-way classification
(smooth, nonsmooth-convex, nonsmooth-concave) is now used directly.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Simplify is_atom_smooth docstring

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Propagate initial values to reduced variables in CvxAttr2Constr

When CvxAttr2Constr creates reduced variables for dim-reducing attributes
(e.g., diag), the original variable's initial value was not being lowered
and assigned to the reduced variable. This caused NLP solvers to fail with
"Variable has no value" during initial point construction.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Handle sparse values when propagating diag variable initials

When a diag variable has its value set as a sparse matrix, extract the
diagonal directly via scipy rather than passing it through np.diag which
does not support sparse inputs.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Revert "Handle sparse values when propagating diag variable initials"

This reverts commit e422565.

* Revert "Propagate initial values to reduced variables in CvxAttr2Constr"

This reverts commit d02a758.

* fix pnorm nonsmooth convex

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

When CvxAttr2Constr creates reduced variables for dimension-reducing
attributes (e.g. diag=True), the initial value was not propagated to
the reduced variable, causing NLP solvers to fail. This mirrors the
existing value propagation already done for parameters.

Also handle sparse diagonal matrices in lower_value() by using
.diagonal() instead of np.diag() which doesn't accept sparse input.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

#153)

* Make is_linearizable_convex/is_linearizable_concave abstract on Expression

Enforce implementation in all Expression subclasses by uncommenting
@abc.abstractmethod decorators. Add missing implementations to indicator
(convex, not concave) and PartialProblem (delegates to is_convex/is_concave).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* remove is_smooth from max

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Separates test-only utility from production code by moving it to
cvxpy/tests/nlp_tests/derivative_checker.py and updating all 19
test file imports.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

* Address review comments: remove redundant nonsmooth methods, add types, docs

- Remove is_atom_nonsmooth_convex/is_atom_nonsmooth_concave from base Atom
  class and 10 atom subclasses; use existing is_atom_convex/is_atom_concave
  in DNLP composition rules instead
- Add type annotations and docstrings to NLPsolver, Bounds, and Oracles
  in nlp_solver.py
- Document Variable.sample_bounds with class-level type annotation and
  docstring
- Revert GENERAL_PROJECTION_TOL back to 1e-10 (was loosened for removed
  IPOPT derivative checker)
- Update CLAUDE.md atom classification docs to reflect simplified API

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Move sample_bounds docs from #: comments into Variable class docstring

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

…156)

* Extract NLP solving logic from problem.py into nlp_solving_chain.py

Move the ~85-line NLP block from Problem._solve() and the two initial
point methods into a dedicated module. This addresses PR review feedback:
- NLP chain building, initial point logic, and solve orchestration now
  live in cvxpy/reductions/solvers/nlp_solving_chain.py
- Use var.get_bounds() instead of var.bounds so sign attributes (nonneg,
  nonpos) are incorporated into bounds automatically
- Initial point helpers are now private module-level functions instead of
  public Problem methods

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Update NLP initial point tests to use _set_nlp_initial_point

Tests now import and call the module-level helper directly instead of
the removed Problem.set_NLP_initial_point() method.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Remove debug print and expand comment in best_of loop

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Gate best_of print on verbose flag

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Address PR review comments on nlp_solving_chain extraction

- Add circular import comment in problem.py explaining the deferred import
- Move NLP_SOLVER_VARIANTS from nlp_solving_chain.py to defines.py
- Set BOUNDED_VARIABLES = True on NLPsolver base class and use it in
  _build_nlp_chain (matching the conic solver pattern in solving_chain.py)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* adds changes to test_problem with parametrize

* Revert verbose unpack comment to original single-line version

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

…160)

* uses new coo handling

* added comment

* fix pre-commit

* fix error message if sample bounds are not set (#161)

* Remove conda from CI, use uv + system IPOPT

Replace conda-based test_nlp_solvers workflow with uv, installing IPOPT
via system packages (apt on Ubuntu, brew on macOS) instead of conda-forge.
Uncomment IPOPT optional dependency in pyproject.toml so uv sync --extra
IPOPT works. Update installation docs in CLAUDE.md and README.md.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Keep IPOPT extra commented out to avoid --all-extras breakage

The test_optional_solvers workflow uses uv sync --all-extras, which would
try to build cyipopt without system IPOPT installed. Instead, install
cyipopt directly via uv pip in test_nlp_solvers where the system library
is available.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Remove UV_SYSTEM_PYTHON to fix externally-managed env error

uv pip install fails on Ubuntu when UV_SYSTEM_PYTHON=1 because the
system Python is externally managed. Removing it lets uv use its own
venv created by uv sync.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Install LAPACK and BLAS dev libraries for cyipopt build on Ubuntu

cyipopt links against LAPACK and BLAS which are not installed by default
on Ubuntu runners.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Use uv venv + uv pip install instead of uv sync for NLP workflow

uv sync manages a locked environment that doesn't play well with
uv pip install for additional packages. Switch to uv venv + uv pip
install to manage the environment directly.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Includes upstream fixes for quad_form canonicalization, gradient handling,
LDL factorization optimization, CI updates, and documentation improvements.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* uses new coo handling

* added comment

* fix pre-commit

* fix error message if sample bounds are not set

* docs for DNLP

* remove whitespace

* split table up into two

* uses new coo handling

* added comment

* fix pre-commit

* fix error message if sample bounds are not set

* docs for DNLP

* remove whitespace

* split table up into two

* cache C problem in best of

- Remove `submodules: recursive` from CI workflows (build, test_nlp_solvers, test_backends)
- Revert `_grad` in affine_atom.py to upstream version
- Remove NaN-allowance block in leaf.py added for NLP structural jacobian
- Update is_dnlp() docstrings to use linearizable terminology
- Add normalize_shape() helper in converters.py, replacing 7 inline occurrences
- Remove unnecessary comment in nlp_solver.py
- Export UNO in cvxpy/__init__.py
- Revert README.md to upstream CVXPY version

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

#168)

* remove unnecessary point in dom + clean up C_problem + clean up oracle

* fix william's comments

…nology (#170)

- Revert whitespace-only changes in huber, abs, ceil, affine_atom, binary_operators
- Revert cosmetic rename and _jacobian_operator addition in index.py
- Revert isnan check in leaf.py back to original error handling
- Restore removed blank line in quad_form.py
- Update DNLP constraint docstrings to use "linearizable convex/concave" terminology

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

- Remove unused `self.broadcast_type` attribute from `broadcast_to.__init__`
  (was previously used for derivative computation, now handled elsewhere)
- Revert CLAUDE.md to upstream master version (DNLP-specific changes were
  broader than intended for the PR)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Extract the DIFFENGINE special-case from ConeMatrixStuffing.apply() and
conic_solver.format_constraints() into a clean DiffengineMatrixStuffing
sibling class. The solving chain now branches at construction time, and
format_constraints uses polymorphic apply_restruct_mat() dispatch.

- Extract lower_and_order_constraints() for shared constraint lowering
- Add DiffengineMatrixStuffing extending MatrixStuffing (new file)
- Add apply_restruct_mat() to ParamConeProg and DiffengineConeProgram
- Add SymbolicQuadForm converter for QP path through diffengine
- Remove DIFFENGINE branch from ConeMatrixStuffing and isinstance check
  from conic_solver.py

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>