Fig. 5: Distribution of database-calculated and experimentally-measured band gap widths for a representative sample of phases. | npj Computational Materials

Fig. 5: Distribution of database-calculated and experimentally-measured band gap widths for a representative sample of phases.

From: Identifying candidate hosts for quantum defects via data mining

Fig. 5

Comparison of reported experimental band gaps to those calculated by a the Materials Project, b AFLOW-ML, c JARVIS-DFT (OPT only), d JARVIS-DFT (OPT-mBJ), e AFLOW (unfit), and f AFLOW (fit). On average, the standard DFT calculations a, c, e and machine-learning predictions b were found to underestimate the experimental band gaps by roughly 40–50%. JARVIS-DFT OPT-mBJ calculations d were found to underestimate measured values by about 18% on average, while AFLOW fitted band gap calculations f were found to overestimate by about 2% on average, with several significant outlying underestimations. The red lines show the linear least squares fits to the data.

Back to article page