TCSP3.0
A template-based crystal structure prediction system that fixes the element mapping deficiencies of previous versions through coefficient-preserving element mapping and synchronous substitution. It integrates enhancement modules including period-number retrieval, soft oxidation-state evidence, radius compatibility scoring, structure deduplication, and machine-learning potential relaxation.
Timeline (last 90 days)
TCSP3.0 enforces coefficient-preserving element mapping, synchronously replaces all species on the parsed structure, and validates the composition and unit cell of each candidate.
Ablation experiments show that most of TCSP3.0's performance improvement on T180 and T800 comes from the correctness fixes (coefficient-preserving mapping and synchronous replacement).
TCSP3.0 introduces enhancement modules such as period-number retrieval, soft oxidation-state evidence, radius compatibility scoring, structure deduplication, ORB-v3 machine learning potential relaxation, and energy ranking.
Ablation experiments show that embedding-based retrieval is crucial for coverage, while the remaining scoring features have small or negligible effects; structure deduplication significantly improves the Top-5 match rate.
On the T180 benchmark, TCSP3.0 achieves Top-1 and Top-20 match rates of 72.2% and 90.6%, respectively, whereas TCSP2.0 reaches only 28.3% and 45.6%.
On the newly constructed 800-target benchmark T800, TCSP3.0 achieves a Top-20 match rate of 74.0%.
On T180, TCSP3.0's Top-1 match rate (72.2%) outperforms PNcsp+ and the deep generative models CrystalFlow and MaskGXT (at most 63.9%).
Even without relaxation, TCSP3.0 achieves Top-1 67.8% and Top-20 90.6% with a median time of only 28 seconds per target on a dual-core CPU, about ten times faster than CrystalFlow.
All Facts (8)
TCSP3.0 enforces coefficient-preserving element mapping, synchronously replaces all species on the parsed structure, and validates the composition and unit cell of each candidate.
50%UnverifiedAblation experiments show that most of TCSP3.0's performance improvement on T180 and T800 comes from the correctness fixes (coefficient-preserving mapping and synchronous replacement).
50%UnverifiedTCSP3.0 introduces enhancement modules such as period-number retrieval, soft oxidation-state evidence, radius compatibility scoring, structure deduplication, ORB-v3 machine learning potential relaxation, and energy ranking.
50%UnverifiedAblation experiments show that embedding-based retrieval is crucial for coverage, while the remaining scoring features have small or negligible effects; structure deduplication significantly improves the Top-5 match rate.
50%UnverifiedOn the T180 benchmark, TCSP3.0 achieves Top-1 and Top-20 match rates of 72.2% and 90.6%, respectively, whereas TCSP2.0 reaches only 28.3% and 45.6%.
50%UnverifiedOn the newly constructed 800-target benchmark T800, TCSP3.0 achieves a Top-20 match rate of 74.0%.
50%UnverifiedOn T180, TCSP3.0's Top-1 match rate (72.2%) outperforms PNcsp+ and the deep generative models CrystalFlow and MaskGXT (at most 63.9%).
50%UnverifiedEven without relaxation, TCSP3.0 achieves Top-1 67.8% and Top-20 90.6% with a median time of only 28 seconds per target on a dual-core CPU, about ten times faster than CrystalFlow.
50%