Introducing MOSOX - A simpler route from model to solution

[Mark Howells]

Dear Open-Minded-Modellers ... 

Introducing MOSOX - A simple route from model to solution

Energy models are often built with tools such as Linopy, Pyomo and JuMP. These are flexible, but they add code, packages and data structures between the equations and the solver.

MOSOX takes a narrower route. It compiles readable close to Maths models (AMPL/GNUMathProg) directly into standard matrices and can solve them with HiGHS, and other solvers. In OSeMOSYS tests, it compiled models up to 6.5 times faster than GLPK’s glpsol and used less peak memory on the largest case.

Linopy’s benchmarks show that it is faster and more memory-efficient than Pyomo, and broadly comparable with JuMP. MOSOX asks a simpler question: when an LP or MIP can be written directly, is a full programming framework needed?

Faster model generation at similar memory use means less computing time and, in general, lower energy use and environmental impact.

A small algebraic language is also easier to learn. The equations remain visible, so students, engineers, economists and domain experts can inspect the model without learning a large software framework. The same property supports AI-assisted modelling: an AI can draft the model while a human reviews every variable, constraint and assumption.

MOSOX is one tile in the emerging MOSAIC-modelling.ai ecosystem. It is free to use and designed to work with open standards and existing tools, not to lock users into a particular model, workflow or software stack.

Keep the model close to the mathematics. Keep the path to the solver short.

Paper: https://www.cambridge.org/engage/coe/article-details/6a4bb6f1810b9dcc82d3cc9f

Sample LP maths:

example.mod:

/* Basketball Lineup Model */ 
set PLAYERS := 1..7; 
 
/* Abilities: Assisting, Throwing, Rebounding, Defense */ 
param A{PLAYERS}; param T{PLAYERS}; param R{PLAYERS}; param D{PLAYERS}; 
 
/* Decision variable: 1 if player plays, 0 otherwise */ 
var y{i in PLAYERS} binary; 
 
/* Objective: Maximize total defensive ability */ 
maximize total_defense: sum{i in PLAYERS} D[i] * y[i]; 
 
/* Constraints */ 
s.t. five_players: sum{i in PLAYERS} y[i] = 5; 
 
/* Position constraints based on player capabilities */ 
s.t. backfield: y[1] + y[3] + y[5] + y[7] >= 3; 
s.t. frontfield: y[4] + y[5] + y[6] + y[7] >= 2; 
s.t. midfield: y[2] + y[3] + y[4] + y[6] >= 1; 
 
/* Average ability constraints (Avg >= 2 across 5 players) */ 
s.t. avg_assist: sum{i in PLAYERS} A[i] * y[i] >= 10; 
s.t. avg_throw:  sum{i in PLAYERS} T[i] * y[i] >= 10; 
s.t. avg_rebound: sum{i in PLAYERS} R[i] * y[i] >= 10; 
 
/* Logical constraints */ 
s.t. exclusivity: y[3] + y[6] <= 1;   /* If 3 plays, 6 cannot */ 
s.t. requirement_4: y[1] <= y[4];      /* If 1 plays, 4 must play */ 
s.t. requirement_5: y[1] <= y[5];      /* If 1 plays, 5 must play */ 
s.t. mandatory_choice: y[2] + y[3] >= 1; /* Either 2 or 3 must play */ 
 
data; 
param A := 1 3 2 2 3 2 1 1 3 2 3 6 3 7 3; 
param T := 1 3 2 1 3 3 4 3 5 3 6 1 7 2; 
param R := 1 1 2 3 3 2 4 3 5 1 6 2 7 2; 
param D := 1 3 2 2 3 2 4 1 5 2 6 3 7 1; 
end;

Enjoy [!]

Best, Mark - and the MOSOX team [!]

Mark Howells (Prof)

Director ClimateCompatibleGrowth.com see a quick video intro here
Joint Director STEER Centre, Loughborough University

Principal Research Fellow, Centre for Environmental Policy, Imperial College

Editor in Chief Energy Strategy Reviews
Google Scholar