A new way to compute — and to earn.
Encode an optimization problem; the market relaxes to the answer, and the trades that get there return a profit — a trading method built for scientists and engineers.
f(x)
Stop predicting. Start solving.
Same market — a completely different game. One asks you to guess. The other asks you to solve.
- ✕Predict where prices goA directional bet on the market.
- ✕Win some, lose someA fragile edge that decays.
- ✕You're left with a positionNo knowledge produced — only risk.
- ✓State a problem to solveMathematics, not a bet.
- ✓The market relaxes to the answerIts equilibrium is the solution.
- ✓Keep the answer — and the profitTwo outputs from one run.
The market does the computing. You read the answer.
Encode
State your problem — an objective, a system, a constraint set — and it becomes structured orders on a live market.
Perturb
Samādhān calibrates the market’s dynamics (θ, σ, μ) and nudges prices away from rest.
Relax
Mean-reversion and arbitrage pull prices to the one configuration that satisfies your problem.
Extract
The fills spell out the solution. The trades along the way return a profit.
Prices follow an Ornstein–Uhlenbeck pull toward equilibrium — physical relaxation.
A residual r = Ax−b becomes a restoring force, zero exactly at the solution.
Arb bots trade away your spreads for profit — correcting constraints, funding the solve.
The hardest problems, run on the market.
Not toy demos — each was encoded and solved with real capital on a live exchange, and recorded with publication-grade results.
E. coli metabolism
23 metabolic fluxes fit to ¹³C-MFA data under 18 mass-balance + 12 thermodynamic constraints.
Rosenbrock 6D
The classic non-convex benchmark, minimized live across six mean-reverting pairs.
Lennard-Jones clusters
Minimum-energy configuration of an atomic cluster — hard global optimization.
Toeplitz eigenvalue
Dominant eigenvalue of a structured 8×8 matrix, read from market relaxation.
Boolean satisfiability
Every clause satisfied at once; each violation a penalty force.
Shortest path
Path-cost constraints push flow onto the minimum-cost route.
Euler ODE
Initial-value problems integrated as the residual is driven to zero.
LLM steering
Steer a model toward a target behaviour with the MCM optimizer.
You leave with the answer — and the return.
Residual 0.15 · confidence 0.70 · verified against the known global minimum of the 6D Rosenbrock function.
Per solve on $1,000 of working capital — PnL stayed positive across the entire 31-minute run. The computation paid for itself.
Put your hardest problem into the market.
Samādhān is in private alpha. Tell us what you want solved.