Goldilocks

Goldilocks

Towards Smarter & Greener Materials Simulations

Overview

Goldilocks is an open ecosystem for efficient, responsible compute in DFT-based materials and molecular science research.

The Challenge

DFT convergence is easy to get wrong. Over-converge and you waste compute; under-converge and the calculation itself is wasted.

Solution

Deterministic, ML-predicted parameters, database lookups to skip redundant runs, and MLIP shortcuts — all orchestrated by an LLM agent.

Outcomes

The Goldilocks Agent, available today — turning structures into ready-to-run calculations, with less compute and carbon wasted.

The online version runs right in your browser on Qwen, with no setup or API key needed. For the full experience, the local version lets you bring your own API key or a local model, and adds extra tools such as the MLIP Playground.

The Goldilocks Agent is under active development, so please double-check results before relying on them.

Developed at UKRI-STFC Scientific Computing Department · Funded by Ada Lovelace Centre EPSRC
The ecosystem

Four modules, one pipeline

Each piece hands off to the next — from raw DFT data, to trained models, to deterministic parameters, to natural-language workflows.

◆
goldilocks-data

Data

Runs real DFT calculations at scale, producing convergence-tested datasets that every downstream model is trained on.

◈
goldilocks-ml

Machine Learning

Trains on those datasets and publishes versioned model artifacts, each with a clearly defined domain of applicability.

◉
goldilocks-core

Core

Turns a structure into DFT parameters, input files, and submission scripts — deterministic, and every value traceable to its source.

◐
goldilocks-agent

Agent

Natural-language front end that orchestrates the pipeline — database lookups, MLIP shortcuts, and post-processing, with plain-language explanations.

The Goldilocks Project

Overview

The Goldilocks project focuses on improving the efficiency and responsible use of computing resources in Density Functional Theory (DFT) calculations. DFT is a crucial method in materials and molecular science, supporting the R&D of functional materials for environmental sustainability.

DFT-based research consumes a significant amount of resources on supercomputers worldwide, including a substantial portion of the UK's national supercomputing service, ARCHER2. As computing power increases, the environmental impact of DFT-based research grows, making it essential to optimize its efficiency.

This project aims to provide practical tools and evidence-based best practices for researchers and the compute-resources distribution chain.

The Challenge

DFT calculations involve numerical approximations that require careful convergence. Inexperienced users may either over-converge, leading to unnecessarily costly calculations, or under-converge, resulting in useless calculations. Both scenarios waste valuable compute resources and electricity. It is estimated that a significant proportion of DFT calculations suffer from such inefficiencies.

Solution

The Goldilocks project will develop a tool and evidence-based best practices to guide researchers in making the optimal ('Goldilocks') choice of convergence parameters. This will be achieved by:

  • Training machine learning (ML) models to predict the convergence parameters for DFT numerical approximations, tailored to the required accuracy in common scientific investigations.
  • Creating a tool applicable across all DFT codes commonly used in the UK.

Expected Outcomes

  • Significantly increased efficiency and assurance levels in the use of UKRI and EPSRC hardware and software infrastructure.
  • Reduction in wasted compute resources, leading to substantial improvements in the environmental sustainability of research infrastructure.
  • Quantification and extrapolation of efficiency gains.
  • Dissemination of best practices globally.
  • Advancement towards ML-based automatic generation of inputs for DFT calculations.
  • Automatic a priori calculation of compute resources and carbon footprint.
  • Democratization of DFT research.

Organisations

Lead Research Organisation: STFC — Laboratories
Department: Scientific Computing Department

The Goldilocks Project

People

Project Lead

Alin-Marin Elena

Alin-Marin Elena

Principal Investigator
Susmita Basak

Susmita Basak

Co-Investigator

Project Members

Former Project Members