Skip to content

Real-time time-dependent density functional theory study of optical Kerr nonlinearities in semiconductors and metals

FieldValue
Journalnpj Computational Materials
Published2026-08-01
AuthorsXiao Chen, Nicolas Tancogne-Dejean, Jingyin Huang, Silvana Botti
DOI10.1038/s41524-026-02237-0

At 4 eV (310 nm), MBJLDA rt-TDDFT gave diamond a two-photon absorption coefficient of 0.2211 cm/GW, compared with 0.2175 cm/GW from experiment. The method also reproduced gold’s increasing Kerr-type third-order susceptibility with pulse duration. Thermal accumulation was identified as a likely cause of overestimated measured nonlinearities in high-repetition-rate experiments.

Real-time method reproduces experimental observables

Section titled “Real-time method reproduces experimental observables”

The calculations used real-space rt-TDDFT in OCTOPUS for bulk diamond, silicon, gold, and iridium. Gaussian pulses at several intensities and durations were applied to each crystal. The authors fitted the resulting effective complex refractive index to extract n₂, β, and the Kerr-type χ(3).

Because all four crystals are centrosymmetric, second-order nonlinear processes vanish. Electronic current served as the basic observable, which captured interband and intraband contributions without a separate Drude term. A self-consistent induced field linked the vacuum-defined laser intensity to the field acting inside the material.

MBJLDA was used for diamond and silicon, and PBEsol for gold and iridium. The calculations used a 0.16 Å mesh and a 20 × 20 × 20 Monkhorst-Pack grid. The optimized lattice constants differed from experimental values by less than 1%.

Benchmarks included z-scan, self-phase modulation, third-harmonic generation, four-wave mixing, and pump-probe transmission measurements. This coverage allowed direct comparison with several quantities inferred by experimental methods.

Diamond response and pulse-duration limits

Section titled “Diamond response and pulse-duration limits”

Pump photon energy was swept from 1.5 to 5 eV, with four intensity points calculated at each energy. The fitted coefficients were compared with diamond z-scan and pump-probe transmission data.

Pump energyMBJLDA n₂ [10⁻¹⁹ m²/W]Experimental n₂ [10⁻¹⁹ m²/W]MBJLDA β [cm/GW]Experimental β [cm/GW]
3 eV, 413 nm1.24300.80.01190.0407
4 eV, 310 nm0.20240.61330.22110.2175

The experimental n₂ at 3 eV was measured at 2.9 eV. Both experimental β values were extracted from polynomial fits.

At 4 eV, the two-photon energy of 8 eV exceeds the MBJLDA direct gap of 6.42 eV, so direct two-photon channels are available. The resulting β was close to experiment. At 3 eV, the two-photon energy of 6 eV lies above the indirect gap of 5.47 eV but below the direct gap of 6.5 eV.

Phonon-assisted transitions dominate the 3 eV regime. The calculations freeze ion positions and omit electron-phonon coupling, so the paper underpredicts β at that energy. Calculated and measured diamond Kerr-type χ(3) data were all of the same order, approximately 10⁻²¹ m²/V².

The production diamond calculations used 10 fs of total propagation and produced an approximately 4 fs FWHM pulse. Published experiments used 100 fs and 120 fs pulses.

A 400 nm pump at 3.1 eV lies far below the 5.56 eV LDA direct gap. Virtual transitions dominate in this regime, and |n₂| changes little with pulse duration. At 250 nm and 5.0 eV, coherence builds closer to resonance, and |n₂| increases for longer pulses. The 250 nm calculation produced a negative n₂, so its magnitude was plotted.

The TPA coefficient showed only minor duration dependence at both photon energies. These results support short-pulse calculations for off-resonant diamond, while near-resonant calculations require longer pulses to capture the coherence buildup.

Silicon exposes bandgap and phonon effects

Section titled “Silicon exposes bandgap and phonon effects”

MBJLDA gives silicon direct and indirect gaps of 3.08 eV and 1.00 eV, close to experimental values of 3.4 eV and 1.12 eV. At 1 eV, MBJLDA gives n₂ of 3.265 × 10⁻¹⁸ m²/W. Experimental values are 1.303 × 10⁻¹⁸ m²/W at 1 eV and 0.932 × 10⁻¹⁸ m²/W at 1.12 eV.

PBE and LDA instead give n₂ values of 6.410 and 7.750 × 10⁻¹⁸ m²/W. Their smaller bandgaps produce larger calculated nonlinearities. The same functional ordering appears for β. MBJLDA gives 0.249 × 10⁻¹¹ m/W, while PBE and LDA give 1.360 and 1.723 × 10⁻¹¹ m/W. Experimental values are 0.536 and 0.628 × 10⁻¹¹ m/W.

Between 0.5 and 1 eV, two-photon absorption in silicon is dominated by phonon-assisted indirect transitions. Freezing the ions underestimates β in that range. Above 1 eV, the reported β switches to the imaginary part of the nonlinear refractive index because single-photon absorption dominates.

These results show why a single functional cannot reproduce the entire experimental dispersion. The calculated bandgap strongly affects the location and magnitude of nonlinear absorption.

Metals require damping and pulse-duration treatment

Section titled “Metals require damping and pulse-duration treatment”

Gold was calculated at 300 nm (4.13 eV) using propagation times of 12, 36, 73, 121, and 242 fs. Four vector-potential amplitudes, 27, 109, 190, and 272 eV, were used at each time. The pulse FWHM was defined as 2T/5.

Adiabatic rt-TDDFT has no explicit dissipation, so numerical errors produced persistent post-pulse oscillations at long times. The authors damped the induced vector potential tail and required its terminal value below 0.27 μeV. They also selected damping to preserve the linear refractive index and resolve the third-harmonic peak.

At 73 fs, the corrected spectrum resolved both the 300 nm fundamental and the third-harmonic peak. After correction, |χ(3)| increased monotonically with pulse duration, consistent with gold experiments.

The authors attribute this trend to delayed hot-electron dynamics. They describe electron-electron scattering, electron-phonon scattering, and later energy exchange with the lattice as processes occurring over tens to hundreds of femtoseconds.

For iridium at 700 nm, bulk PBEsol gives n₂ of −0.90 × 10⁻¹⁵ m²/W. Nanoparticle measurements report 1.69 × 10⁻¹⁵ m²/W. The sign difference agrees with effective-medium results for bulk-limit iridium, where the local-field phase can change the sign. The authors relate the smaller bulk magnitude mainly to the shorter simulated pulse.

Thermal history and model limits affect interpretation

Section titled “Thermal history and model limits affect interpretation”

Heat accumulation can distort measured nonlinearities when the pulse interval is shorter than t_c = W²/(4D). At 300 K, the listed diffusion coefficients are 12.90 cm²/s for diamond, 0.88 cm²/s for silicon, and 1.27 cm²/s for gold.

For diamond, the authors estimate t_c around 3 μs, although a common-material estimate exceeds 40 μs. A 1 MHz pulse train leaves only 1 μs between pulses, which can cause heat accumulation and systematic overestimation of pure TPA β. The 1 kHz experiments leave 1000 μs for diffusion between pulses.

Silicon’s lower D implies a longer diffusion time. Its 500 Hz measurements report lower n₂ and β than the 1 kHz measurements. An 80 MHz silicon-core-fiber measurement can remain consistent with 1 kHz wafer data because confinement reduces W and t_c.

The metal damping acts as a numerical proxy for relaxation mechanisms absent from adiabatic rt-TDDFT. At higher intensities, fifth- and seventh-order responses appear beyond the fitted third-order regime, but the paper does not analyze them. Excitonic corrections are expected to be small off resonance, while a systematic real-time Bethe-Salpeter study remains outstanding.

Implications for nonlinear optical material selection

Section titled “Implications for nonlinear optical material selection”

Engineers selecting diamond for Kerr applications should use wavelength-resolved n₂ and β rather than a single coefficient. The 4 eV result shows that direct two-photon absorption can be captured well. The 3 eV result shows that phonon-assisted absorption requires a treatment beyond frozen-ion rt-TDDFT.

Agreement in β did not establish agreement in n₂ at 4 eV. Both coefficients therefore require separate validation against the target wavelength and intensity. Silicon offers a larger Kerr coefficient than diamond, but its smaller gap changes the absorption channel and introduces phonon-assisted transitions.

Metals require calculations matched to the experimental pulse duration or an explicit relaxation treatment. Experimental protocols should also record repetition rate, beam diameter, and pulse duration. Those parameters determine whether the reported coefficient describes intrinsic nonlinear optics or accumulated heating.

For the paper’s diamond Kerr application, a sample must provide a centrosymmetric bulk response under controlled wavelength, intensity, and pulse duration. 6CCVD’s diamond materials catalog lists MPCVD single-crystal and polycrystalline diamond plates and wafers up to 125 mm. Its engineering support can help translate the modeled optical conditions into a specimen specification.

View Original Abstract

Abstract We employ real-time time-dependent density functional theory (rt-TDDFT) to calculate third-order nonlinear optical properties in solids, focusing on the optical Kerr effect: the Kerr coefficient n 2 , the two-photon absorption coefficient β , and the Kerr-type third-order susceptibility χ (3) (− ω ∣ ω , ω , − ω ). We consider centrosymmetric bulk materials, including semiconductors (diamond, silicon) and metals (gold, iridium), and benchmark our results against available experimental data. For semiconductors, we find good agreement with experiment and show that the nonlinear response is robust with respect to pulse duration for photon energies below the bandgap. For metals, we demonstrate that ∣ χ (3) ∣ grows monotonically with pulse duration, consistent with experimental observations, and attribute this behavior to delayed contributions from hot-electron dynamics. We further show that thermal effects in high-repetition-rate experiments can lead to significant overestimation of the measured nonlinearities. Our work establishes rt-TDDFT as a reliable first-principles framework for modeling nonlinear optical phenomena across diverse material classes, including metals, going beyond perturbative approaches by naturally capturing intraband dynamics and pulse-duration effects.