Claude-Assisted Computation Sets New Precision Record for Mandelbrot Set Area

Geoffrey Irving announced that, with Claude's help in analysis and deep CUDA optimization, he has pushed the estimate of the Mandelbrot set's area to μ(M) = 1.506591883653 ± 4.7e-11 (95% confidence interval), a 60x precision improvement over Thorstenson's 2012 record. This is another notable example of Claude participating in nontrivial scientific computation and is worth attention.

Confirmed

• Method framework: A quadtree proves cells are entirely inside the set (Newton's method to find attracting cycles, Koebe theorem to bound distances) or entirely outside (distance estimation), leaving roughly 1e13 boundary-straddling cells for random sampling, with each point iterated up to 2^32 steps; total compute was a few H200 GPU-days.
• Handling slow boundary orbits: Long orbits are subsampled quadratically to keep the estimate unbiased, then the tiny fraction (about 6e-10) needing more than 2^32 steps is extrapolated; the extrapolation was backtested on held-out data and directly validated out to 2^36 steps.
• Open source and verification: The work builds on the author's prior repository girving/mandelbrot, which bounds the set's area from above via Böttcher series; the new estimate lives on the hybrid-area branch. Double-precision results were cross-checked against double-double, with independent runs agreeing within ±2.7e-10 at the 95% confidence level.
• Comparison with Hsing Lo's 2025 estimate of 1.5065918902 ± 5.4e-9: that value sits 6.5e-9 above the new result, just outside its confidence interval. The author reproduced Lo's membership tests, found the bias small and mostly pushing downward, and concluded the gap looks like statistical fluctuation.

Why it matters

• The result improves the precision of the Mandelbrot set's area—a long-standing open numerical problem—by more than an order of magnitude, with fully open, reproducible methods and code.
• Claude played a substantive role throughout the analysis and CUDA optimization, demonstrating the feasibility of AI-assisted high-precision scientific computing.

2026-10-06 ~ 2026-10-06 · 5 related posts

• [source] Claude-assisted CUDA compute pins Mandelbrot set area to 1.506591883653, 60x tighter than 2012 record — geoffreyirving · 2026-10-06
• [source] How the new Mandelbrot area estimate works: quadtree pruning plus random boundary sampling on H200s — geoffreyirving · 2026-10-06
• Subsampling and extrapolation keep the Mandelbrot area estimate unbiased near the boundary — geoffreyirving · 2026-10-06
• [source] Mandelbrot area estimate builds on Böttcher-series upper bound, open-sourced with cross-checked precision — geoffreyirving · 2026-10-06
• New estimate sits 6.5e-9 below Hsing Lo's 2025 value; reproduction suggests the gap is a fluctuation — geoffreyirving · 2026-10-06