Update arxiv

README.md

@@ -5,7 +5,7 @@
 [![pre-commit](https://img.shields.io/badge/pre--commit-enabled-brightgreen?logo=pre-commit&logoColor=white)](https://github.com/pre-commit/pre-commit)
 [![Documentation](https://img.shields.io/badge/docs-readthedocs-blue?logo=readthedocs)](https://furax-cs.readthedocs.io/en/latest/)
 [![Results Explorer](https://img.shields.io/badge/%F0%9F%A4%97%20Results-Explorer-yellow?)](https://askabalan-furax-cs-results.hf.space/)
-[![arXiv](https://img.shields.io/badge/arXiv-XXXX.XXXXX-b31b1b?logo=arxiv)](https://arxiv.org/abs/XXXX.XXXXX)
+[![arXiv](https://img.shields.io/badge/arXiv-2604.08463-b31b1b?logo=arxiv)](https://arxiv.org/abs/2604.08463)
 
 **FURAX-CS** (FURAX Component Separation) is a Python package designed to benchmark and implement advanced component separation techniques for Cosmic Microwave Background (CMB) analysis. It leverages **JAX** for high-performance computing on GPUs and implements novel adaptive clustering methods.
 

docs/api/index.md

@@ -18,12 +18,15 @@
 
 ## Optimization
 
+Optimization is provided by the [CADRE](https://github.com/CMBSciPol/CADRE) package.
+See the [minimization docs](../minimization.md) for full solver reference.
+
 ```{eval-rst}
-.. autofunction:: furax_cs.optim.minimize.minimize
+.. autofunction:: cadre.minimize.minimize
 ```
 
 ```{eval-rst}
-.. autofunction:: furax_cs.optim.solvers.get_solver
+.. autofunction:: cadre.solvers.get_solver
 ```
 
 ## Binning

docs/minimization.md

@@ -1,6 +1,8 @@
 # Minimization Solvers
 
-`furax-cs` provides a unified interface for various minimization solvers, including wrappers for **Optax**, **Optimistix**, and **SciPy**.
+`furax-cs` exposes a unified interface for various minimization solvers via the
+[**CADRE**](https://github.com/CMBSciPol/CADRE) package (Constraint-Aware Descent Routine Executor),
+which provides the underlying implementations for **Optax**, **Optimistix**, and **SciPy** solvers.
 
 ## Available Solvers
 
@@ -161,7 +163,7 @@ All other solvers (`optax_lbfgs`, `adam`, `optimistix_*`, etc.) benefit from ext
 
 ### Single Run
 ```python
-from furax_cs import minimize
+from cadre import minimize
 
 final_params, state = minimize(
     fn=my_loss_fn,
@@ -183,7 +185,7 @@ Here is an example of running the same optimization problem with multiple solver
 import jax
 import jax.numpy as jnp
 import optimistix as optx
-from furax_cs.optim.solvers import get_solver
+from cadre.solvers import get_solver
 from functools import partial
 
 # Define a simple quadratic loss function

docs/quick_start.md

@@ -36,7 +36,8 @@ from furax.obs import (
 from furax.obs.stokes import Stokes
 from jax_healpy.clustering import get_cutout_from_mask, get_fullmap_from_cutout
 
-from furax_cs import generate_noise_operator, kmeans_clusters, minimize
+from cadre import minimize
+from furax_cs import generate_noise_operator, kmeans_clusters
 from furax_cs.data import (
     get_instrument,
     get_mask,

notebooks/06_Optimizer_Benchmark.ipynb

@@ -6,7 +6,7 @@
    "source": [
     "# Optimizer Benchmark\n",
     "\n",
-    "This notebook compares different optimization solvers from `furax_cs.optim` on the Rosenbrock function.\n",
+    "This notebook compares different optimization solvers from `cadre` on the Rosenbrock function.\n",
     "\n",
     "## Test Matrix\n",
     "\n",
@@ -45,7 +45,7 @@
     "import jax.numpy as jnp\n",
     "import matplotlib.pyplot as plt\n",
     "import pandas as pd\n",
-    "from furax_cs.optim import minimize\n",
+    "from cadre import minimize\n",
     "\n",
     "# Enable 64-bit precision\n",
     "jax.config.update(\"jax_enable_x64\", True)"
@@ -196,28 +196,6 @@
     "plt.show()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 13,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "Array([  35.178079  ,  839.05471775,  946.67697787, 1124.11267378,\n",
-       "       1344.27794419, 1591.60078999, 2023.58745027, 2241.40393647,\n",
-       "       2429.79396429, 2553.55842245], dtype=float64)"
-      ]
-     },
-     "execution_count": 13,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "eigvals"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 36,
@@ -255,32 +233,6 @@
     "plt.show()"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 32,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "Array([[  2.,   0.,   0., ...,   0.,   0.,   0.],\n",
-       "       [  0., 202.,   0., ...,   0.,   0.,   0.],\n",
-       "       [  0.,   0., 202., ...,   0.,   0.,   0.],\n",
-       "       ...,\n",
-       "       [  0.,   0.,   0., ..., 202.,   0.,   0.],\n",
-       "       [  0.,   0.,   0., ...,   0., 202.,   0.],\n",
-       "       [  0.,   0.,   0., ...,   0.,   0., 200.]], dtype=float64)"
-      ]
-     },
-     "execution_count": 32,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "H"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 13,

pyproject.toml

@@ -17,8 +17,7 @@ dependencies = [
   "pysm3",
   "astropy",
   "optimistix",
-  "cobyqa",
-  "jaxopt",
+  "jax-cadre",
   "scienceplots"
 ]
 description = "GPU Powered CMB Parametric Component Seperation using Furax and JAX"
@@ -30,7 +29,7 @@ version = "0.1.1"
 
 [project.optional-dependencies]
 all = [
-  "furax-cs[plotting,dev,benchmarks,io,docs]"
+  "furax-cs[plotting,dev,benchmarks,io,docs,scipy]"
 ]
 benchmarks = ["jax-hpc-profiler>=v0.3.2"]
 dev = [
@@ -53,6 +52,9 @@ plotting = [
   "seaborn",
   "scienceplots"
 ]
+scipy = [
+  "jax-cadre[scipy]"
+]
 
 [project.scripts]
 bench-bcp = "furax_cs.scripts.bench_bcp:main"

slurm/runners/11-binning.sh

@@ -69,7 +69,8 @@ RESULTS_DIR="RESULTS/KMEANS_C1D1S1"
 OUTPUT_DIR="RESULTS/BINNING_C1D1S1"
 
 # 3 optimal runs (same as section_45.sh)
-RUNS=(-r 'BD7500_TD500_BS500_GAL020' 'BD10000_TD6000_BS300_GAL040' 'BD10000_TD7000_BS500_GAL060')
+RUNS=(-r 'BD7000_TD500_BS500_GAL020' 'BD10000_TD500_BS150_GAL040' 'BD10000_TD2500_BS300_GAL060')
+
 
 # =============================================================================
 # Phase 1 & 2: For each bin config, bin then run kmeans

src/furax_cs/__init__.py

@@ -2,6 +2,18 @@
 
 from importlib import metadata
 
+from cadre import (
+    SOLVER_NAMES,
+    ScipyMinimizeState,
+    apply_projection,
+    condition,
+    get_solver,
+    lbfgs_backtrack,
+    lbfgs_zoom,
+    minimize,
+    scipy_minimize,
+)
+
 from . import r_analysis
 from .binning import bin_parameter_map
 from .data import (
@@ -28,17 +40,6 @@
 from .kmeans_clusters import kmeans_clusters
 from .multires_clusters import multires_clusters
 from .noise import generate_noise_operator
-from .optim import (
-    SOLVER_NAMES,
-    ScipyMinimizeState,
-    apply_projection,
-    condition,
-    get_solver,
-    lbfgs_backtrack,
-    lbfgs_zoom,
-    minimize,
-    scipy_minimize,
-)
 
 __all__ = [
     "bin_parameter_map",

src/furax_cs/optim/__init__.py

@@ -1,41 +1,8 @@
-"""
-Optimization utilities for FURAX component separation.
-
-This package provides:
-- L-BFGS solvers with zoom and backtracking linesearch
-- Box projection transformation for constrained optimization
-- Unified optimization interface supporting optax, optimistix, and scipy
-- Function conditioning (parameter transformation and gradient-based scaling)
+"""Compatibility shim — optimization has moved to the cadre package.
 
-Example usage:
-    >>> from furax_cs.optim import minimize
-    >>>
-    >>> # Simple optimization
-    >>> params, state = minimize(
-    ...     fn=objective,
-    ...     init_params={'beta': 1.5},
-    ...     solver_name='optax_lbfgs',
-    ... )
+Import from cadre directly:
+    from cadre import minimize, get_solver, condition, ...
 """
 
-from .minimize import ScipyMinimizeState, minimize, scipy_minimize
-from .solvers import (
-    SOLVER_NAMES,
-    apply_projection,
-    get_solver,
-    lbfgs_backtrack,
-    lbfgs_zoom,
-)
-from .utils import condition
-
-__all__ = [
-    "SOLVER_NAMES",
-    "ScipyMinimizeState",
-    "apply_projection",
-    "condition",
-    "get_solver",
-    "lbfgs_backtrack",
-    "lbfgs_zoom",
-    "scipy_minimize",
-    "minimize",
-]
+from cadre import *  # noqa: F401, F403
+from cadre import __all__  # noqa: F401