Full case study: /case-studies/gcn-optical-transitions

Observable

Electronic structure and optical response (band structure, DOS, dielectric function, oscillator strengths) computed for g-C₃N₄ with transition-metal doping (Pt, Pd, Co) in periodic models.

Claim

Metal doping shifts absorption onset and activates forbidden optical transitions via symmetry breaking. Assignments to specific defect sites require structural identification beyond computation alone.

Load-Bearing Constraints

Axiomatic

  • Optical transitions obey symmetry and selection rules
  • Breaking symmetry redistributes oscillator strength but doesn’t create unique spectral fingerprints

Measurement

  • Experimental spectra average over polymerization, stacking, grain boundaries, defects
  • No atomic-scale structural probes available

Fabrication

  • Synthesis: variable C/N stoichiometry, incomplete polymerization, residual precursors
  • Metal dopants: multiple local environments (coordination, oxidation state)
  • Sample-to-sample variation in dopant concentration and distribution

Statistical

  • Finite sampling: ~10-20 arrangements per dopant
  • Rare but optically dominant coordination environments may be absent

Computational

  • PBE: underestimates gaps by ~1 eV (affects absolute energies, less so relative trends)
  • Oscillator strengths: sensitive to functional choice and k-point sampling
  • Supercell size: 72-108 atoms (limits dopant separation modeling)
  • Calculated response is for specific structural model, not “the material”

Primary Limiting Factor

Structural model non-uniqueness.

Multiple doping sites, stacking variants, and defect arrangements produce absorption onsets within 0.2-0.3 eV. Cannot assign experimental peaks to computed structures without independent structural validation.

What This Ruled Out

  • One-to-one peak assignments to specific dopant sites (computation alone insufficient)
  • Treating computed spectrum as synthesis-independent property
  • Quantitative experimental transition energy prediction (PBE error ~1 eV)

What Remains Non-Identifiable

  • Which microscopic motif (or ensemble) dominates measured spectra
  • Dopant coordination and oxidation state (need X-ray absorption spectroscopy)
  • Whether spectral shifts arise from electronic structure vs. morphology/disorder
  • Contributions from edge states, grain boundaries, structural defects

What Would Help

  • Broader configuration sampling with uncertainty quantification
  • Higher-level theory (HSE06, GW-BSE) for representative subset
  • Structural characterization (EXAFS, XANES) to constrain local environment
  • Controlled synthesis varying only dopant concentration
  • Optical measurements on single-crystal or epitaxial samples (reduce ensemble averaging)

Methods Referenced

Related constraints: Alloy sampling (similar finite sampling issues)

Analysis date: Spring 2024 - Present
My experience: Second computational project, learning excited-state modeling and optical response from DFT

This project taught me the difference between qualitative mechanisms (symmetry breaking—robust) and quantitative predictions (specific peak positions—not robust with PBE). I initially thought more DFT would solve the assignment problem. It doesn’t—you need structural information from experiment.

Back to Constraints Full Case Study