3Kaiserberge

Engineering & Research

Solver technology that works where standard methods fail.

Request Benchmark

Convergio — Open Source Convergence Detection

Your solver finished 3000 iterations ago. Convergio tells you when to stop — and when something is wrong.

from convergio import detect

result = detect(residual_history, residual_target=1e-8)
# returns: "converged" · "stalling" · "oscillating" · "diverging"

500+

Installs

6

States

MIT

License

CAP — Canonical Attractor Preconditioner

15 of 16 industry matrices solved. Performance benchmarks on SuiteSparse collection.

15/16

Solved

10

Rescued

84x

Peak Speedup

1.2M

Max DOF

Matrix Baseline Solver CAP Engine (v0.8) Result
nos4CG (84 iter)1 iter84x SPEEDUP
nos6CG (666 iter)12 iter55x SPEEDUP
ex5CG FAIL2 iterRESCUED
offshoreCG FAIL (40.2s)2.8sRESCUED
cavity10GMRES FAIL84 iterRESCUED
cavity16GMRES FAIL173 iterRESCUED
bcsstk14CG FAIL58 iterRESCUED
bcsstk15CG FAIL74 iterRESCUED
bcsstk17CG FAIL389 iter [strong]RESCUED
bcsstk18CG FAIL184 iterRESCUED
msc10848CG FAIL1845 iter [strong]RESCUED
s3rmt3m3CG FAIL1313 iter [strong]RESCUED
thermal2CG (83s)71s14% FASTER
apache2CG (39s)34s13% FASTER
thermal1CG (1345 iter)122 iter15% SLOWER
ex11CG OKFAILNon-optimal case

Walltime measured single-threaded on commodity hardware. CAP is not parallelized — these numbers are a lower bound.

Zero external dependencies · Header-only C++ · Integrates into any solver stack.

Request Evaluation
Convergio Enterprise

C++ Header-Only. Drop into any C++/Fortran simulation.

  • · Single header: convergio.hpp
  • · Zero dependencies, zero config
  • · Real-time: watch() inside your solver loop
  • · Post-hoc: detect() on completed runs
  • · Adaptive regime detection
Request Evaluation