Innovation Hub

Convex Optimization Tool Enables Rapid and Accurate Portfolio Management at NASA with Global Optimal Solutions
Number coding


Each year, the National Aeronautics and Space Administration (NASA) Small Business Innovation Research (SBIR) program receives approximately 1,800 proposals and must select only around 350 for awards. The vast range of possible selection combinations results in thousands of possible topic/subtopic portfolios, creating an intractable solution space for probabilistic optimization methodologies. Furthermore, the value of a given portfolio is subject to multiple, often conflicting, scoring schemas, adding to the complexity of the selection process.


To make the selection solution space more manageable, REI developed a principled convex optimization tool capable of calculating portfolios quickly (80 ms) and guaranteeing global optimality. The algorithm’s efficient performance allows for the generation of entire Pareto surfaces, where different scoring methodologies can be compared against one another. While the example demonstrates a comparison between two possible scoring schemas, generalizing to n-dimensions is feasible.

REI employed a principled convex optimization approach that ensures guaranteed convergence at the global optima and exhibits known, predictable convergence characteristics despite the presence of nearly 850 parameters. Moreover, the conic formulation and its associated solver rely solely on open-source technologies, enhancing solution portability across various environments.


The primary advantages of using the convex optimization methodology developed by REI include significant time savings and increased confidence in decision-making. This is achieved by swiftly comparing numerous portfolios across previously unattainable axes of measurement. In comparison to previous approaches, the convex optimization methodology explored by REI offers a faster solution for evaluating and making informed decisions.

Capabilities Shown

  • Performant Algorithm Analysis
  • Principal Convex Optimization Methodology
  • Mixed-Integer Second-Order Conic Programing (MISOCP)