Fix: align SplitBregman docs and cost with implementation, fix tol=0

[mrava87]

Closes #808.

Docs

• The effective weight of each L1 term is epsRL1s[i]**2: epsRL1s[i] is used both as the weight of the augmented L2 term in the x-subproblem and as the threshold of the shrinkage step. The cost function in the Notes, the constrained/split formulation, and the epsRL1s parameter descriptions now reflect this.
• Use d_i for the split variable (was y_i, clashing with the data y), and add tau to the Bregman update.
• Describe the x-subproblem solvers precisely (scipy.sparse.linalg.lsqr for engine="scipy" with NumPy arrays, cgls for engine="pylops" or other array types) and the closed-form shrinkage of the d-subproblem, including the stopping criterion.
• Fix other docstring errors: duplicated nregsL1 attribute (should be nregsL2), types of b/d, OMP/IRLS copy-paste in step/run, size of the initial guess returned by setup, engine/show_inner/kwargs_lsqr descriptions mentioning lsqr only, over-indented itershow/preallocate in splitbregman, "Split-Bergman" typo.

Code

• The recorded cost now matches the documented cost function: 1/2 on the L2 terms and epsRL1s**2 on the L1 terms (previously epsRL2s without 1/2, and unit weight on L1). The solution is unchanged.
• tol=0 no longer skips all outer iterations: the model-update norm is initialized to inf, so at least one iteration is always run.

Tests

• test_SplitBregman_tol0: all niter_outer iterations are run with tol=0.
• test_SplitBregman_cost: cost[0] matches the documented cost computed by hand (with both L1 and L2 terms and nonzero dataregsL2).

🤖 Generated with Claude Code

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 0 complexity · 0 duplication

Metric	Results
Complexity	0
Duplication	0

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}