Clarabel Conic Interior Point Solver for Rust / Python

Warnings

• gotcha Clarabel's default solver settings are optimized for 64-bit float data. Using 32-bit floats (e.g., `np.float32`) may lead to suboptimal performance or numerical inaccuracies unless solver tolerances are explicitly relaxed.
  Fix: Ensure problem data (P, q, A, b) is provided in 64-bit float format (`np.float64`). If 32-bit precision is required, adjust solver settings like `settings.tol_gap_abs` and `settings.tol_gap_rel` accordingly.
• gotcha When updating problem data (e.g., in iterative algorithms), modifications to the `cones` array are not permitted. Additionally, data updates are disallowed if `settings.chordal_decomposition_enable = True` or `settings.presolve_enable = True`.
  Fix: If the cone structure or presolve/chordal decomposition settings need to change, a new solver instance must be created. For allowed data updates, ensure the sparsity pattern and problem dimensions remain constant.
• gotcha Clarabel expects sparse matrix data (P and A) to be in Compressed Sparse Column (CSC) format, as produced by `scipy.sparse`. For the quadratic objective matrix `P`, only its upper triangular part should be provided in CSC format.
  Fix: Always convert sparse matrices to `scipy.sparse.csc_matrix`. For the 'P' matrix, use `P = sparse.triu(P).tocsc()` to ensure correct format and content.
• breaking From Clarabel version 0.11.0 onwards, the version string is accessed via `clarabel.__version__` (an attribute) instead of `clarabel.__version__()` (a method).
  Fix: Update any code that retrieves the Clarabel version from `clarabel.__version__()` to `clarabel.__version__`.
• deprecated Starting with CVXPY 1.5, Clarabel has become the default interior-point solver, replacing ECOS. CVXPY 1.6 will entirely remove ECOS as a dependency. Users relying on ECOS's specific behavior via CVXPY might experience regressions.
  Fix: If experiencing issues with Clarabel as the default in CVXPY, explicitly specify a different solver (e.g., `problem.solve(solver='ECOS')` if ECOS is installed, or `solver='SCS'`). Consider installing ECOS manually if needed with CVXPY >=1.6.

Install

• pip install clarabel Install stable version

Imports

• clarabel

  import clarabel

• numpy

  import numpy as np

• sparse

  from scipy import sparse

• DefaultSolver

  clarabel.DefaultSolver

  Classes like DefaultSolver, DefaultSettings, and cone types are typically accessed via the top-level 'clarabel' module after `import clarabel`.
• __version__

  clarabel.__version__

  Prior to version 0.11.0, `__version__` was a callable method. It is now a direct attribute.

Quickstart

This quickstart demonstrates how to define and solve a simple quadratic program (QP) using Clarabel. It shows how to define the objective function matrix (P) and vector (q), constraint matrix (A) and vector (b), and the cone structure (K). The problem data is expected in `numpy` arrays and `scipy.sparse.csc_matrix` format.

import clarabel
import numpy as np
from scipy import sparse

# Define problem data for a simple QP: minimize 0.5 * x.T * P * x + q.T * x
# subject to A * x + s = b, s in K

P = sparse.csc_matrix([[3., 1., -1.], [1., 4., 2.], [-1., 2., 5.]])
P = sparse.triu(P).tocsc() # Clarabel expects upper triangular part in CSC format
q = np.array([1., 2., -3.])

A = sparse.csc_matrix([
    [1., 1., -1.], # Equality constraint (ZeroConeT)
    [0., 1., 0.],  # Inequality constraint (NonnegativeConeT)
    [0., 0., 1.],  # Inequality constraint (NonnegativeConeT)
    [0., 0., 0.]   # Placeholder for SOC (SecondOrderConeT)
])

b = np.array([1., 2., 2., 0.])

# Define cone constraints: ZeroConeT(dim), NonnegativeConeT(dim), SecondOrderConeT(dim)
cones = [clarabel.ZeroConeT(1), clarabel.NonnegativeConeT(2), clarabel.SecondOrderConeT(1)]

# Instantiate and solve the solver
settings = clarabel.DefaultSettings()
settings.verbose = False # Disable verbose output
solver = clarabel.DefaultSolver(P, q, A, b, cones, settings)
solver.solve()

# Print results
print(f"Solution x: {solver.solution.x}")
print(f"Objective value: {solver.solution.obj_val}")
print(f"Solver status: {solver.solution.status}")