This is a short tutorial on how to run linear-response TDDFT computations for absorption and emission spectroscopy. The TDDFT module enables a description of excitation energies and excited-state computations within the Tamm-Dancoff approximation (TDA) featuring GGA and hybrid functionals as well as semi-empirical simplified TDA kernels. The details of the implementation can be found in https://aip.scitation.org/doi/full/10.1063/1.5078682 (Ref. [1]) and in https://pubs.acs.org/doi/10.1021/acs.jctc.2c00144 for corresponding excited-state gradients. Note that the current module is based on an earlier TDDFT implementation https://chimia.ch/chimia/article/view/2005_499. Please cite these papers if you were to use the TDDFT module for the computation of excitation energies or excited-state gradients.

The implementation in CP2K is based on the Tamm-Dancoff approximation (TDA), which describes each excited state $p$ with the excitation energy $\Omega^p$ and the corresponding excited-state eigenvectors $\mathbf{X}^p$ as an Hermitian eigenvalue problem

\begin{equation} \label{tda_equation} \begin{aligned} \mathbf{A} \mathbf{X}^p = \Omega^p \mathbf{X}^p \, , \\ \sum_{\kappa k} [ F_{\mu \kappa \sigma} \delta_{ik} - F_{ik \sigma} S_{\mu \kappa} ] X^p_{\kappa k \sigma} + \sum_{\lambda} K_{\mu \lambda \sigma} [\mathbf{X}^p] C_{\lambda i \sigma} = \sum_{\kappa} \Omega^p S_{\mu \kappa} X^p_{\kappa i \sigma} \, . \end{aligned} \end{equation}

The Hermitian matrix $\mathbf{A}$ contains as zeroth-order contributions the difference in the Kohn-Sham (KS) orbital energies $\mathbf{F}$, and to first order kernel contributions $\mathbf{K}$ which comprise –depending on the chosen density functional approximation – Coulomb $\mathbf{J}$ and exact exchange $\mathbf{K}^{\rm{\tiny{EX}}}$ contributions as well as contributions due to the exchange-correlation (XC) potential $\mathbf{V}^{\rm{\tiny{XC}}}$ and kernel $\mathbf{f}^{\rm{\tiny{XC}}}$,

\begin{equation} \label{f_and_k_matrix} \begin{aligned} F_{\mu \nu \sigma} [\mathbf{D}] &= h_{\mu \nu} + J_{\mu \nu \sigma} [\mathbf{D}] - a_{\rm{\tiny{EX}}}K^{\rm{\tiny{EX}}}_{\mu \nu \sigma} [\mathbf{D}] + V_{\mu \nu \sigma}^{\rm{\tiny{XC}}} \, , \\ K_{\mu \nu \sigma} [\mathbf{D}^{\rm{\tiny{X}}}] &= J_{\mu \nu \sigma} [\mathbf{D}^{\rm{\tiny{X}}}] - a_{\rm{\tiny{EX}}} K^{\rm{\tiny{EX}}}_{\mu \nu \sigma}[\mathbf{D}^{\rm{\tiny{X}}}] + \sum_{\kappa \lambda \sigma'} f^{\rm{\tiny{XC}}}_{\mu \nu \sigma,\kappa \lambda \sigma'} D_{\kappa \lambda \sigma'}^{\rm{\tiny{X}}} \, . \end{aligned} \end{equation}

$\mathbf{S}$ denotes the atomic-orbital overlap matrix, $\mathbf{C}$ the occupied ground-state KS orbitals and $\mathbf{D}$ and $\mathbf{D}^{\rm{\tiny{X}}}$ ground-state and response density matrices,

\begin{equation}\label{density_matrices} \begin{aligned} D_{\mu \nu \sigma} = \sum_k C_{\mu k \sigma} C_{\nu k \sigma}^{\rm{T}} \, , \\ D_{\mu \nu \sigma}^{\rm{\tiny{X}}} = \frac{1}{2} \sum_{k} ( X^p_{\mu k \sigma} C_{\nu k \sigma}^{\rm{T}} + C_{\mu k \sigma} (X^p_{\nu k \sigma})^{\rm{T}} ) \, . \end{aligned} \end{equation}

Within the current implementation, symmetrization and orthogonalization of the response density matrix is ensured at each step of the Davidson algorithm. The current implementation features to approximate the exact exchange contribution of hybrid functionals using the auxiliary density matrix method (ADMM). Furthermore, the standard kernel can be approximated using the semi-empirical simplified Tamm-Dancoff approximation (sTDA), neglecting in this case XC contributions and approximating both Coulomb and exchange contributions $\mathbf{J}$ and $\mathbf{K}$ using semi-empirical operators $\boldsymbol{\gamma}^{\rm{\tiny{J}}}$ and $\boldsymbol{\gamma}^{\rm{\tiny{K}}}$ depending on the interatomic distance $R_{AB}$ of atoms $A$ and $B$,

\begin{equation} \label{stda_kernel} \begin{aligned} \gamma^{\rm{\tiny{J}}}(A,B) &= \left ( \frac{1}{(R_{AB})^{\alpha} + \eta^{-\alpha}} \right)^{1/\alpha} \, , \\ \gamma^{\rm{\tiny{K}}}(A,B) &= \left ( \frac{1}{(R_{AB})^{\beta} +( a_{\rm{\tiny{EX}}}\eta)^{- \beta} } \right )^{1/\beta} \, , \end{aligned} \end{equation}

that depend on the chemical hardness $\eta$, the Fock-exchange mixing parameter $a_{\rm{\tiny{EX}}}$ and powers of $\alpha$ and $\beta$ for either Coulomb and exchange interactions.

Within the current implementation, oscillator strengths can be calculated for molecular systems in the length form and for periodic systems using the velocity form (see Ref.[1]).

Based on Eq.\ (\ref{tda_equation}), excited-state gradients can be formulated based on a variational Lagrangian for each excited state $p$,

\begin{equation} \begin{aligned} L [\mathbf{X}, \mathbf{C}, \Omega, \bar{\mathbf{W}}^{\rm{\tiny{X}}}, \bar{\mathbf{Z}}, \bar{\mathbf{W}}^{\rm{\tiny{C}}} ] &= \Omega - \sum_{\kappa \lambda k l \sigma} \Omega ( X_{\kappa k \sigma }^{\rm{T}} S_{\kappa \lambda } X_{\lambda l \sigma} - \delta_{kl} ) \\ &- \sum_{kl \sigma} ( \bar{W}_{kl \sigma}^{\rm{\tiny{X}}} )^{\rm{T}} \sum_{\kappa \lambda} \frac{1}{2} ( C_{\kappa k \sigma}^{\rm{T}} S_{\kappa \lambda} X_{\lambda l \sigma} + X_{\kappa k \sigma}^{\rm{T}}S_{\kappa \lambda} C_{\lambda l \sigma}) \\ &+ \sum_{\kappa k \sigma}( \bar{Z}_{\kappa k \sigma})^{\rm{T}} \sum_{\lambda} ( F_{\kappa \lambda \sigma}C_{\lambda k \sigma} - S_{\kappa \lambda } C_{\lambda k \sigma} \varepsilon_{k \sigma}) \\ &- \sum_{kl\sigma} (\bar{W}^{\rm{\tiny{C}}}_{kl \sigma})^{\rm{T}} ( S_{kl \sigma} - \delta_{kl})\, . \end{aligned} \end{equation}

introducing Lagrange multipliers $\bar{\mathbf{W}}^{\rm{\tiny{X}}}$, $\bar{\mathbf{W}}^{\rm{\tiny{C}}}$, and $\bar{\mathbf{Z}}$ to ensure stationarity of the corresponding ground-state (GS) equations and to account for the geometric dependence of the Gaussian orbitals and thus requiring to solve the Z vector equation iteratively.

To compute absorption spectra, parameters defining the LR-TDDFT computation have to be specified in the TDDFPT subsection of the section PROPERTIES of section FORCE_EVAL. Furthermore, RUN_TYPE has to be set to ENERGY and the underlying KS ground-state reference has to be specified in the DFT section.

The most important keywords and subsections of TDDFPT are:

• KERNEL: option for the kernel matrix $\mathbf{K}$ to choose between the full kernel for GGA or hybrid functionals and the simplified TDA kernel
• NSTATES: number of excitation energies to be computed
• CONVERGENCE: threshold for the convergence of the Davidson algorithm
• RKS_TRIPLETS: option to switch from the default computation of singlet excitation energies to triplet excitation energies

To compute excited-state gradients and thus corresponding fluorescence spectra, the excited state to be optimized furthermore has to be specified by adding the subsection EXCITED_STATES of the section ''DFT.