This fork extends CVXPY to provide explicit control of warm start and update behavior for OSQP, SCS, QPALM, and PROXQP. More solvers may be supported in the future.

Version baseline:

• This modified fork is developed from CVXPY v1.9.0.

In original CVXPY, warm start and update behavior is mostly tied to a single warm_start flag and internal cache behavior. This is limiting when:

1. Solving many problems with similar structure, where reusing a small pool of Problem instances is preferred over storing every instance.
2. Needing to control update and warm start independently, or warm-start from user-provided values (not only from the previous solve).

The original CVXPY supports warm start from the previous solve and update, both from the internal cached data. I modify osqp_qpif. It supports lower-level control by passing extra data to the solver. An example is stored in test_new.py.

The table below summarizes the new options for controlling OSQP warm start and update behavior:

OSQP update semantics in this implementation:

• If update=True, a cached solver instance must already exist from a previous solve on the same Problem object.
• update=True updates solver data (q/l/u, and P/A values when changed) without rebuilding from scratch.
• If warm_start=True, warm_start_solution_dict must contain both x and y with valid lengths.
• If warm_start=False, the solver is explicitly warm-started from zeros.

The original CVXPY only supports warm start from the previous solve, by the internal cached data and no update is supported. I modify scs_conif.

The table below summarizes the new options for controlling SCS warm start and update behavior:

SCS update semantics in this implementation:

• If update=True, a cached solver instance must already exist from a previous solve on the same Problem object.
• Current update path calls solver.update(b=..., c=...) (it does not update A/P structure).
• If warm_start=True, warm_start_solution_dict accepts keys in {x, y, s} with valid lengths.
• If warm_start=False, the solver runs without user-provided warm-start vectors.
• For SCS v2, the custom mode keeps the acceleration_lookback=0 retry behavior when update=False and no explicit acceleration_lookback is provided.

The original CVXPY supports warm start from cached solver state for QPALM. I modify qpalm_qpif to expose separate custom controls for update and warm start.

The table below summarizes the new options for controlling QPALM warm start and update behavior:

QPALM update semantics in this implementation:

• If update=True, a cached solver instance must already exist from a previous solve on the same Problem object.
• update=True updates QPALM problem data (Q/A, q, and bounds) without rebuilding the solver object when possible.
• If warm_start=True, warm_start_solution_dict must contain both x and y with valid lengths.
• If warm_start=False, the solver is explicitly warm-started from zeros.

The original CVXPY supports warm start from cached solver state for PROXQP. I modify proxqp_qpif to expose separate custom controls for update and warm start.

The table below summarizes the new options for controlling PROXQP warm start and update behavior:

PROXQP update semantics in this implementation:

• If update=True, a cached solver instance must already exist from a previous solve on the same Problem object.
• update=True updates solver data (q, b, bounds, and P/A/F when changed) without rebuilding the solver object when possible.
• If warm_start=True, warm_start_solution_dict must contain x, y, and z with valid lengths.
• If warm_start=False, the solver is explicitly warm-started from zeros.

NOTE: In this custom mode, include both update and warm_start in the data dict. NOTE: data['warm_start'] controls warm start behavior. The warm_start argument passed to chain.solve_via_data(...) is ignored.

You may refer to the following minimal example usage (see test_new.py for more detail):

Because solver warm start operates on canonical (standard-form) data, this feature is implemented at the low level through get_problem_data and solve_via_data.

# prob is the predefined problem instance

# Compile
data, chain, inverse_data = prob.get_problem_data(solver = cp.OSQP)

# Set new options
data['update'] = True
data['warm_start'] = True
data['warm_start_solution_dict'] = {'x': previous_x, 'y': previous_y}

results = chain.solve_via_data(problem=prob, data=data, warm_start=False/True, verbose=False, solver_opts={'polishing': False})

NOTE: The original warm_start flag is ignored.

See test_new.py for concrete usage patterns.

Test files updated in this fork:

• test_new.py demonstrates end-to-end low-level usage for OSQP, SCS, QPALM, and PROXQP.
• cvxpy/tests/test_qp_solvers.py includes OSQP, QPALM, and PROXQP custom-path regression tests.
• cvxpy/tests/test_conic_solvers.py includes SCS custom-path regression tests.

pip install git+https://github.com/xuwkk/cvxpy.git

or

pip install git+https://github.com/xuwkk/cvxpy.git@v1.9.0-lapso.5

For development, clone and install in development mode:

git clone https://github.com/xuwkk/cvxpy.git
pip install -e .

After installation, use CVXPY as usual:

import cvxpy as cp

NOTE: You should uninstall the original CVXPY installation before installing this modified version.

Below is the original README.md from the CVXPY repository.

The CVXPY documentation is at cvxpy.org.

We are building a CVXPY community on Discord. Join the conversation! For issues and long-form discussions, use Github Issues and Github Discussions.

Contents

• Installation
• Getting started
• Issues
• Community
• Contributing
• Team
• Citing

CVXPY is a Python-embedded modeling language for convex optimization problems. It allows you to express your problem in a natural way that follows the math, rather than in the restrictive standard form required by solvers.

For example, the following code solves a least-squares problem where the variable is constrained by lower and upper bounds:

import cvxpy as cp
import numpy

# Problem data.
m = 30
n = 20
numpy.random.seed(1)
A = numpy.random.randn(m, n)
b = numpy.random.randn(m)

# Construct the problem.
x = cp.Variable(n)
objective = cp.Minimize(cp.sum_squares(A @ x - b))
constraints = [0 <= x, x <= 1]
prob = cp.Problem(objective, constraints)

# The optimal objective is returned by prob.solve().
result = prob.solve()
# The optimal value for x is stored in x.value.
print(x.value)
# The optimal Lagrange multiplier for a constraint
# is stored in constraint.dual_value.
print(constraints[0].dual_value)

With CVXPY, you can model

• convex optimization problems,
• mixed-integer convex optimization problems,
• geometric programs, and
• quasiconvex programs.

CVXPY is not a solver. It relies upon the open source solvers Clarabel, SCS, OSQP and HiGHS. Additional solvers are available, but must be installed separately.

CVXPY began as a Stanford University research project. It is now developed by many people, across many institutions and countries.

CVXPY is available on PyPI, and can be installed with

pip install cvxpy

CVXPY can also be installed with conda, using

conda install -c conda-forge cvxpy

CVXPY has the following dependencies:

• Python >= 3.11
• Clarabel >= 0.5.0
• OSQP >= 1.0.0
• SCS >= 3.2.4.post1
• NumPy >= 2.0.0
• SciPy >= 1.13.0
• highspy >= 1.11.0

For detailed instructions, see the installation guide.

To get started with CVXPY, check out the following:

• official CVXPY tutorial
• example library
• API reference

We encourage you to report issues using the Github tracker. We welcome all kinds of issues, especially those related to correctness, documentation, performance, and feature requests.

For basic usage questions (e.g., "Why isn't my problem DCP?"), please use StackOverflow instead.

The CVXPY community consists of researchers, data scientists, software engineers, and students from all over the world. We welcome you to join us!

• To chat with the CVXPY community in real-time, join us on Discord.
• To have longer, in-depth discussions with the CVXPY community, use Github Discussions.
• To share feature requests and bug reports, use Github Issues.

Please be respectful in your communications with the CVXPY community, and make sure to abide by our code of conduct.

We appreciate all contributions. You don't need to be an expert in convex optimization to help out.

You should first install CVXPY from source. Here are some simple ways to start contributing immediately:

• Read the CVXPY source code and improve the documentation, or address TODOs
• Enhance the website documentation
• Browse the issue tracker, and look for issues tagged as "help wanted"
• Polish the example library
• Add a benchmark

If you'd like to add a new example to our library, or implement a new feature, please get in touch with us first to make sure that your priorities align with ours.

Contributions should be submitted as pull requests. A member of the CVXPY development team will review the pull request and guide you through the contributing process.

Before starting work on your contribution, please read the contributing guide.

CVXPY is a community project, built from the contributions of many researchers and engineers.

CVXPY is developed and maintained by Steven Diamond, Akshay Agrawal, Riley Murray, Philipp Schiele, Bartolomeo Stellato, and Parth Nobel, with many others contributing significantly. A non-exhaustive list of people who have shaped CVXPY over the years includes Stephen Boyd, Eric Chu, Robin Verschueren, Jaehyun Park, Enzo Busseti, AJ Friend, Judson Wilson, Chris Dembia, and William Zhang.

For more information about the team and our processes, see our governance document.

If you use CVXPY for academic work, we encourage you to cite our papers. If you use CVXPY in industry, we'd love to hear from you as well, on Discord or over email.