#fts #solver #optimization-solver #fts-solver

fts-solver

A reference solver for flow trading optimization

5 unstable releases

Uses new Rust 2024

new 0.3.0 Apr 15, 2025
0.2.0 Apr 11, 2025
0.1.2 Mar 21, 2025
0.1.1 Mar 21, 2025
0.1.0 Mar 21, 2025

#382 in Algorithms

Download history 337/week @ 2025-03-19 29/week @ 2025-03-26 14/week @ 2025-04-02 205/week @ 2025-04-09

585 downloads per month
Used in 4 crates

MIT license

64KB
1K SLoC

crates.io version docs.rs documentation crates.io downloads crates.io license getting started

Flow Trading Service (FTS)

This crate is part of a collection of crates that together implement flow trading as proposed by Budish, Cramton, et al, in which trade occurs continuously over time via regularly-scheduled batch auctions.

The different crates in this workspace are as follows:

  • fts_core: Defines a set of data primitives and operations but defers the implementations of these operations, consistent with a so-called "hexagonal architecture" approach to separating responsibilities.
  • fts_solver: Provides a reference solver for the flow trading quadratic program.
  • fts_server: A REST API HTTP server for interacting with the solver and persisting state across auctions.
  • fts_demo: An implementation of the core data operations using SQLite, suitable for exploration of flow trading-based marketplaces such as a forward market.

FTS Solver

This package defines a few basic types and a solver interface to operate over these types. Presently, the following solvers are provided:

  • feature = ["clarabel"] -- Uses the Clarabel interior point solver for the quadratic program
  • feature = ["osqp"] -- Uses the OSQP ADMM solver for the quadratic program

Additional solvers will be developed as needed. The present implementations are intended as "reference" for future work.

There are a few additional features exposed by this crate. If an application intends to (de)serialize the primitive data types directly, enabling feature = ["serde"] will provide Serde bindings.

Primitive Types

There are two externally-defined types ProductId and AuthId, which allow the application host to provide their own implementations. These are black-boxes as far as the solver is concerned -- they just need to implement Clone + Eq + Hash + Ord.

This crate defines a Submission<AuthId, ProductId> type, which is intended to encapsulate a single bidder's submission. A submission is a combination of auths and costs: an auth defines a portfolio (a sparse vector over product space) and the minimum and maximum allowable trade of this portfolio. Costs define a linear combination of auths (a group) and a utility function, whose domain additionally constrains the space of feasible outcomes. It is assumed by the solver that auth ids are globally unique; that is, the auth ids should be disjointly partitioned amongst the submissions. (It does not otherwise cause an error, but will likely yield unexpected results.) Refer to the implementations of src/types/auth.rs and src/types/cost.rs for more details on these types. Refer to tests/simple_solve.rs for an example assembling two submissions and solving them.

TODO

  • Warm-start interface
  • Large-scale tests
  • Enhanced dual reporting
  • Automatic determination of error tolerances based on input
  • Bespoke ADMM implementation

Dependencies

~3.5–5MB
~91K SLoC