Math Optimization Examples

The math benchmarks contain 14 optimization problems from the AlphaEvolve paper and additional geometric challenges. These demonstrate how SkyDiscover can discover novel solutions to open mathematical problems.

Circle Packing

Pack 26 circles in a unit square to maximize the sum of their radii. This is problem B.12 from the AlphaEvolve paper.

Initial Program

The starting solution places circles in concentric rings:

# EVOLVE-BLOCK-START
"""Constructor-based circle packing for n=26 circles"""
import numpy as np

def construct_packing():
    """
    Construct a specific arrangement of 26 circles in a unit square
    that attempts to maximize the sum of their radii.

    Returns:
        Tuple of (centers, radii, sum_of_radii)
    """
    n = 26
    centers = np.zeros((n, 2))

    # Place a large circle in the center
    centers[0] = [0.5, 0.5]

    # Place 8 circles around it in a ring
    for i in range(8):
        angle = 2 * np.pi * i / 8
        centers[i + 1] = [0.5 + 0.3 * np.cos(angle), 
                          0.5 + 0.3 * np.sin(angle)]

    # Place 16 more circles in an outer ring
    for i in range(16):
        angle = 2 * np.pi * i / 16
        centers[i + 9] = [0.5 + 0.7 * np.cos(angle), 
                          0.5 + 0.7 * np.sin(angle)]

    # Clip to ensure everything is inside the unit square
    centers = np.clip(centers, 0.01, 0.99)

    # Compute maximum valid radii for this configuration
    radii = compute_max_radii(centers)
    sum_radii = np.sum(radii)

    return centers, radii, sum_radii

def compute_max_radii(centers):
    """Compute maximum radii without overlap"""
    n = centers.shape[0]
    radii = np.ones(n)

    # Limit by distance to square borders
    for i in range(n):
        x, y = centers[i]
        radii[i] = min(x, y, 1 - x, 1 - y)

    # Limit by distance to other circles
    for i in range(n):
        for j in range(i + 1, n):
            dist = np.sqrt(np.sum((centers[i] - centers[j]) ** 2))
            if radii[i] + radii[j] > dist:
                scale = dist / (radii[i] + radii[j])
                radii[i] *= scale
                radii[j] *= scale

    return radii
# EVOLVE-BLOCK-END

Evaluator

The evaluator validates the packing and scores it against the AlphaEvolve benchmark:

import numpy as np

def validate_packing(centers, radii):
    """Validate circles don't overlap and stay in unit square"""
    n = centers.shape[0]
    
    # Check for NaN values
    if np.isnan(centers).any() or np.isnan(radii).any():
        return False
    
    # Check circles are inside the unit square
    for i in range(n):
        x, y = centers[i]
        r = radii[i]
        if x - r < -1e-6 or x + r > 1 + 1e-6 or \
           y - r < -1e-6 or y + r > 1 + 1e-6:
            return False
    
    # Check for overlaps
    for i in range(n):
        for j in range(i + 1, n):
            dist = np.sqrt(np.sum((centers[i] - centers[j]) ** 2))
            if dist < radii[i] + radii[j] - 1e-6:
                return False
    
    return True

def evaluate(program_path):
    """Evaluate circle packing solution"""
    TARGET_VALUE = 2.635  # AlphaEvolve result for n=26
    
    # Import and run the program
    spec = importlib.util.spec_from_file_location("program", program_path)
    program = importlib.util.module_from_spec(spec)
    spec.loader.exec_module(program)
    
    centers, radii, sum_radii = program.run_packing()
    
    # Validate solution
    valid = validate_packing(centers, radii)
    sum_radii = np.sum(radii) if valid else 0.0
    
    # Score relative to target
    target_ratio = sum_radii / TARGET_VALUE if valid else 0.0
    combined_score = target_ratio
    
    return {
        "sum_radii": float(sum_radii),
        "target_ratio": float(target_ratio),
        "validity": 1.0 if valid else 0.0,
        "combined_score": float(combined_score)
    }

Running the Example

cd benchmarks/math/circle_packing

uv run skydiscover-run \
  initial_program.py \
  evaluator.py \
  -c config.yaml \
  -s adaevolve \
  -i 100

Expected Results: AlphaEvolve achieved a sum of radii of 2.635. Your evolved solution should approach or exceed this value.

Heilbronn Triangle

Arrange 11 points inside an equilateral triangle to maximize the area of the smallest triangle formed by any three points.

Initial Program

# EVOLVE-BLOCK-START
import numpy as np

def heilbronn_triangle11() -> np.ndarray:
    """
    Construct an arrangement of 11 points on or inside an equilateral 
    triangle to maximize the area of the smallest triangle formed.

    Returns:
        points: np.ndarray of shape (11, 2) with x,y coordinates
    """
    n = 11
    points = np.zeros((n, 2))
    # Evolution will improve this initial configuration
    return points
# EVOLVE-BLOCK-END

Evaluator

import numpy as np
import itertools

BENCHMARK = 0.036529889880030156  # Target value
NUM_POINTS = 11

def check_inside_triangle(points: np.ndarray):
    """Check points are inside equilateral triangle"""
    # Triangle vertices: (0,0), (1,0), (0.5, sqrt(3)/2)
    for x, y in points:
        cond1 = y >= -1e-6
        cond2 = np.sqrt(3) * x <= np.sqrt(3) - y + 1e-6
        cond3 = y <= np.sqrt(3) * x + 1e-6
        
        if not (cond1 and cond2 and cond3):
            raise ValueError(f"Point ({x}, {y}) outside triangle")

def triangle_area(a: np.array, b: np.array, c: np.array) -> float:
    """Compute area of triangle formed by three points"""
    return abs(a[0] * (b[1] - c[1]) + 
               b[0] * (c[1] - a[1]) + 
               c[0] * (a[1] - b[1])) / 2

def evaluate(program_path: str):
    # Import program and get points
    program = __import__(module_name)
    points = program.heilbronn_triangle11()
    
    # Validate
    if points.shape != (NUM_POINTS, 2):
        raise ValueError(f"Invalid shape: {points.shape}")
    
    check_inside_triangle(points)
    
    # Compute minimum triangle area
    min_area = min(
        triangle_area(p1, p2, p3) 
        for p1, p2, p3 in itertools.combinations(points, 3)
    )
    
    # Normalize by container triangle area
    container_area = triangle_area(
        np.array([0, 0]), 
        np.array([1, 0]), 
        np.array([0.5, np.sqrt(3)/2])
    )
    min_area_normalized = min_area / container_area
    
    return {
        "min_area_normalized": float(min_area_normalized),
        "combined_score": float(min_area_normalized / BENCHMARK)
    }

Running the Example

cd benchmarks/math/heilbronn_triangle

uv run skydiscover-run \
  initial_program.py \
  evaluator.py \
  -c config.yaml \
  -s adaevolve \
  -i 100

All Math Benchmarks

Signal Processing

Path: benchmarks/math/signal_processing/Real-time adaptive filtering for non-stationary time series. Evolve filter coefficients and adaptation strategies.

Autocorrelation Inequalities

Paths:

benchmarks/math/first_autocorr_ineq/ (B.1)
benchmarks/math/second_autocorr_ineq/ (B.2)
benchmarks/math/third_autocorr_ineq/ (B.3)

Upper and lower bounds on autoconvolution constants from AlphaEvolve Appendix B.

Uncertainty Inequality

Path: benchmarks/math/uncertainty_ineq/ (B.4)Upper bound on Fourier uncertainty constant.

Erdos Minimum Overlap

Path: benchmarks/math/erdos_min_overlap/ (B.5)Upper bound on Erdos minimum overlap constant.

Sums/Differences of Finite Sets

Path: benchmarks/math/sums_diffs_finite_sets/ (B.6)Lower bound on the size of sums and differences of finite sets.

Hexagon Packing

Paths:

benchmarks/math/hexagon_packing/11/ (n=11)
benchmarks/math/hexagon_packing/12/ (n=12)

Pack unit hexagons in a regular hexagon (B.7).

Max/Min Distance

Path: benchmarks/math/minimizing_max_min_dist/Minimize the ratio of maximum to minimum pairwise distances (B.8).

Heilbronn Convex

Paths:

benchmarks/math/heilbronn_convex/13/ (n=13)
benchmarks/math/heilbronn_convex/14/ (n=14)

Heilbronn problem for convex regions (B.10).

Circle Packing in Rectangle

Path: benchmarks/math/circle_packing_rect/Pack circles in a rectangle of perimeter 4 (B.13).

Matrix Multiplication

Path: benchmarks/math/matmul/Faster algorithms for matrix multiplication (Appendix A).

Tips for Math Benchmarks

Start Simple

Begin with simple geometric patterns. Evolution will refine them.

Use Validation

Always validate solutions satisfy constraints before scoring.

Normalize Scores

Compare against known benchmarks (like AlphaEvolve results).

Handle Edge Cases

Check for NaN values, invalid shapes, and numerical instability.

Next Steps

Systems Examples

Explore systems optimization

Algorithm Examples

See competitive programming

Create Custom

Build your own benchmark

​Circle Packing

​Initial Program

​Evaluator

​Running the Example

​Heilbronn Triangle

​Initial Program

​Evaluator

​Running the Example

​All Math Benchmarks

​Tips for Math Benchmarks

Start Simple

Use Validation

Normalize Scores

Handle Edge Cases

​Next Steps

Systems Examples

Algorithm Examples

Create Custom

Circle Packing

Initial Program

Evaluator

Running the Example

Heilbronn Triangle

Initial Program

Evaluator

Running the Example

All Math Benchmarks

Tips for Math Benchmarks

Next Steps