saikrishna0817/markov-cero

How saikrishna0817/markov-cero works

markov-cero is a C++ mathematical optimization solver, with CLI, Python, and optional authenticated web entry points. Its main workflow is to parse and classify an LP/MPS model, prepare it for solving, dispatch to an appropriate optimization engine, then verify and report the result; supported MILP cases can also export replayable proofs. The README documents this pipeline and the major LP, MILP, QP, nonlinear, GPU, and ML capabilities. Some internal links below follow that documentation because the corresponding implementation was not sampled.

Main entry points: api.cpp Solve API · primal_verifier.cpp Result verification · engine_milp.cpp MILP branch and cut · server.py Web solve adapter

Components 22

User interfaces

CLI tools apps/markov_cero_solve.cpp
Python binding python/src/bindings.cpp
Web workspace web/src/App.tsx
Web solve adapter web/backend/server.py

Model preparation

Solve API src/api/api.cpp
LP MPS NLP parsers src/io/mps.cpp
Sparse model src/model/model.cpp
Model classification src/model/classifier.cpp
Presolve scaling canonicalize src/presolve/presolve.cpp

Solve engines

LP engines src/api/engine_lp.cpp
MILP branch and cut src/api/engine_milp.cpp
Parallel tree search src/api/engine_parallel.cpp
QP ADMM KKT src/api/engine_qp.cpp
Nonlinear SQP OA src/api/engine_nonlinear.cpp

Verification and results

Result verification src/verify/primal_verifier.cpp
MIP proof replay src/verify/mip_proof.cpp
JSON results apps/json_output.hpp

Specialized capabilities

GPU kernels gpu/src/pdhg_solver.cpp
ML branching scorer src/milp/ml_branching/onnx_scorer.cpp
Refinery modeling src/refinery/refinery_model.cpp

Other components

Model user
Supabase auth

Connections 28

Model user → CLI tools: submits model
Model user → Python binding: submits model
Model user → Web workspace: uses workspace
Web workspace → Web solve adapter: sends solve request
Web solve adapter → Supabase auth: validates session
Web solve adapter → CLI tools: runs solver
CLI tools → Solve API: calls solve
Python binding → Solve API: exposes solve API
Solve API → LP MPS NLP parsers: reads model
LP MPS NLP parsers → Sparse model: builds model
Solve API → Model classification: classifies model
Solve API → Presolve scaling canonicalize: prepares model
Solve API → LP engines: dispatches LP
Solve API → MILP branch and cut: dispatches MILP
Solve API → Parallel tree search: dispatches parallel search
Solve API → QP ADMM KKT: dispatches QP
Solve API → Nonlinear SQP OA: dispatches nonlinear
LP engines → Result verification: checks witnesses
MILP branch and cut → Result verification: checks primal
Parallel tree search → Result verification: checks primal
QP ADMM KKT → Result verification: checks KKT
Nonlinear SQP OA → Result verification: checks local KKT
MILP branch and cut → MIP proof replay: certifies result
Parallel tree search → MIP proof replay: certifies result
Solve API → JSON results: formats result
LP engines → GPU kernels: uses optional backend
MILP branch and cut → ML branching scorer: scores branches
Model user → Refinery modeling: models refinery decisions

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·