Reproducible scientific simulations using ideas from distributed ledger technology

Abstract

Ensuring the reproducibility of scientific simulations is a persistent challenge, despite current best practices like version control and containerization. Factors such as floating-point arithmetic variations, hardware differences, and concurrency issues often prevent bit-for-bit replication of results. This paper investigates the techniques that distributed ledger technologies employ to achieve deterministic computations and application of these techniques to enhance the reproducibility, trustworthiness and verifiability of scientific simulations. We explore two primary approaches: executing simulations directly “on-chain” for complete transparency and deterministic replay, and performing computations “off-chain” while anchoring their integrity to a blockchain via cryptographic proofs, such as Zero-Knowledge Proofs (ZKPs) and Merkle trees.