United StatesTechnology and telecom2026 – ongoingSchrödinger
AlphaEvolve optimization of molecular simulation code
Schrödinger accelerated force-field training and inference 4x with AI-generated code, Google reported
Problem
Computational bottlenecks slowed Schrödinger's machine-learned force fields for molecular simulation. Ewald summation code relied on slow loops.
Approach
Schrödinger worked with Google Cloud to apply AlphaEvolve to its scientific code. The agent generated a batched Ewald summation implementation using parallel matrix multiplication. Candidate programs had to pass correctness and regression tests as well as run faster.
Results
Takeaway
The reported gain concerned force-field computation. It did not measure a corresponding reduction in the full drug-development cycle.
Categories
- Industry
- Technology and telecom
- Business function
- IT and R&D
- Innovation type
- Process innovation
- AI technology
- Generative AIAI agents
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
Google Cloud described AlphaEvolve's optimization of Schrödinger's molecular simulation algorithms. It reported a 4x training and inference speedup after generated code passed scientific correctness checks.
Full details are in the original sources.