A DFT Study on the Excited Electronic States of Cyanopolyynes: Benchmarks and Applications

Highly unsaturated chain molecules are interesting due to their potential application as nanowires and occurrence in interstellar space. Here, we focus on predicting the electronic spectra of polyynic nitriles HC2m+1N (m = 0–13) and dinitriles NC2n+2N (n = 0–14). The results of time-dependent density functional theory (TD-DFT) calculations are compared with the available gas-phase and noble gas matrix experimental data. We assessed the performance of fifteen functionals and five basis sets for reproducing (i) vibrationless electronic excitation energies and (ii) vibrational frequencies in the singlet excited states. We found that the basis sets of at least triple-ζ quality were necessary to describe the long molecules with alternate single and triple bonds. Vibrational frequency scaling factors are similar for the ground and excited states. The benchmarked spectroscopic parameters were shown to be acceptably reproduced with adequately chosen functionals, in particular ωB97X, CAM-B3LYP, B3LYP, B971, and B972. Select functionals were applied to study the electronic excitation of molecules up to HC27N and C30N2. It is demonstrated that optical excitation leads to a shift from the polyyne- to a cumulene-like electronic structure.


Introduction
One of the challenges of modern quantum chemistry is to develop an approach that would make the predictions on the excited electronic states simple, reliable, and relatively cheap. The most precise results are obtained by ab initio electronic structure methods, which are computationally demanding. Conversely, a popular, inexpensive method that certainly lacks the accuracy of advanced ab initio approaches is the timedependent density functional theory (TD-DFT). It assumes that electronic density slowly varies over time, allowing for the direct application of density functionals developed for ground-state calculations [1]. That approximation, the adopted linear response formalism, and the lack of the exact ground state functional are the main deficiencies of TD-DFT. Additionally, the use of extended orbital basis sets is usually impossible. As a result, proper testing of DFT methods is crucial for their subsequent application.
It may seem that a statistical comparison of the calculation results with experimental data should not pose any problems. However, in the case of electronic transitions, where both the electronic and vibrational contributions shape the spectrum, selecting the appropriate reference data is not trivial. Large-scale testing projects are constrained by the necessity of relatively time-consuming optimization of excited state molecular geometries. Additionally, a large part of experimental data come from the measurement of solutions, and the solvent always modifies the spectrum, sometimes very significantly. TD DFT performance benchmarks obtained for the gas-phase atomic excitation energies [2], where experimental data are well known, are hardly transferrable to molecular species, in particular the polyatomic ones. Alternatively, reference data for vertical electronic excitations of the molecules can be obtained ab initio [3][4][5] and then used to test various DFT methods [6], although the adequacy of theoretically derived reference databases may be a In this work, we computed the electronic excitation energies for long (n up to 14, m up to 13) CPYs, including those that could not, to date, have been measured. We preceded these predictions with benchmark studies on some shorter species, in order to select the most promising functional for that class of molecules. Additionally, scaling factors for vibrational frequencies in the excited electronic states could be estimated here for the first time.

Electronic Excitation Energy
For a given state , we define the percent error between the theoretical and experimental excitation energy as:

Electronic Excitation Energy
For a given state i, we define the percent error between the theoretical E t i and experimental E e i excitation energy as: For the whole training set encompassing the data for n excited states, we define: 1. mean percent error: mean absolute percent error: Tables 1 and 2 list these statistical parameters for our TD-DFT computations, while Figure 2 shows the S 1 -S 0 and S 2 -S 0 excitation energies derived with select functionals, together with the respective experimental values. There, regardless of the functional and basis set, the computed singlet-singlet excitation energies are lower by about 10% than the experimental ones. Underestimation of electronic excitation energy by TD-B3LYP was already reported for some polyynes [56] and for a variety of extended π-electronic systems [57]. This observation is rather generally applicable to TD DFT predictions, at least for the appropriately selected basis sets; similar conclusions were drawn from our study on atomic excitation energies [2] and from the comparison of TD-DFT results with high-quality ab initio benchmark data [6]. As anticipated, the accuracy of calculations grew with the size of the applied orbital basis set (i.e., we did not observe any error-cancelation effects that would produce an opposite trend). The smallest Dunning-type basis set, aug-cc-pVDZ, tended to provide slightly bent structures (rather than the linear ones), in disagreement with the results derived with more extensive basis sets and, most importantly, contradicting the available experimental data. In these cases, forcing a linear symmetry led to the appearance of imaginary frequencies. Consequently, we do not recommend the use of aug-cc-pVDZ in the calculations of excited electronic states of polyynic chains. On the other hand, no statistically meaningful difference appeared between the triple-ζ quality basis sets aug-cc-pVTZ and def2-TZVPD. The above remarks concerning the influence of the orbital basis set on the quality of TD-DFT-derived results do not entirely agree with what was recently reported by the Head-Gordon's group [6]. We associate that discrepancy with entirely different training sets and the nature of the test. Combining the observations from both studies, we can safely recommend the use of aug-cc-pVTZ in TD-DFT analyses of excited state potential energy surfaces. Table 1. Mean percent errors (with respect to the measured gas-phase values) of the singlet-singlet excitation energy calculations performed with various functional/basis set combinations. ∆ is defined by Equation (2). See Table S1 for ∆ st , and ∆ abs (defined by Equations (3) and (4)).   Table S2 for ∆ st , and ∆ abs (defined by Equations (3) and (4)).   Table  S2 for ∆ , and ∆ (defined by Equations (3) and (4)).     (Tables S4-S85).

aug-cc-
The ωB97X significantly outperforms (for the S 1 -S 0 and S 2 -S 0 excitation) the other tested functionals regardless of the applied training set (Figures 3 and 4). However, the ωB97XD, CAM-B3LYP, and BMK seem to be the reasonable choices as well. The good performance of CAM-B3LYP, BMK, and of ωB97-family functionals was expected based on the previous works [2,6,58]. The biggest surprise is the performance of M06-2X, which seems to be good when compared with the results from Kr-matrix and bad when compared with gas-phase data. Matrix-induced shifts of 0-0 transitions are usually much smaller than the difference between the experimental and TD-DFT results. Here, however, M06-2X describes the S 2 states (∆ symmetry) perfectly well, but it encounters a massive problem with the S 1 states (Σ − symmetry). An equivocal behavior of M06-2X stems from the number of S 1 states being small in the Kr training set. M06-2X is a hybrid-meta GGA functional, of high rank in the DFT Jacob's ladder, performing well in other recent benchmark computations [6,10]. That case demonstrates that an a priori estimation of the accuracy of TD-DFT excitation energy is challenging and, regardless of global tests for various molecules and states, it is advisable to perform specifically addressed benchmark studies.
on the previous works [2,6,58]. The biggest surprise is the performance of M06-2X, which seems to be good when compared with the results from Kr-matrix and bad when compared with gas-phase data. Matrix-induced shifts of 0-0 transitions are usually much smaller than the difference between the experimental and TD-DFT results. Here, however, M06-2X describes the S2 states (Δ symmetry) perfectly well, but it encounters a massive problem with the S1 states (Σ − symmetry). An equivocal behavior of M06-2X stems from the number of S1 states being small in the Kr training set. M06-2X is a hybrid-meta GGA functional, of high rank in the DFT Jacob's ladder, performing well in other recent benchmark computations [6,10]. That case demonstrates that an a priori estimation of the accuracy of TD-DFT excitation energy is challenging and, regardless of global tests for various molecules and states, it is advisable to perform specifically addressed benchmark studies.   performance of CAM-B3LYP, BMK, and of ωB97-family functionals was expected based on the previous works [2,6,58]. The biggest surprise is the performance of M06-2X, which seems to be good when compared with the results from Kr-matrix and bad when compared with gas-phase data. Matrix-induced shifts of 0-0 transitions are usually much smaller than the difference between the experimental and TD-DFT results. Here, however, M06-2X describes the S2 states (Δ symmetry) perfectly well, but it encounters a massive problem with the S1 states (Σ − symmetry). An equivocal behavior of M06-2X stems from the number of S1 states being small in the Kr training set. M06-2X is a hybrid-meta GGA functional, of high rank in the DFT Jacob's ladder, performing well in other recent benchmark computations [6,10]. That case demonstrates that an a priori estimation of the accuracy of TD-DFT excitation energy is challenging and, regardless of global tests for various molecules and states, it is advisable to perform specifically addressed benchmark studies.   The above-revealed hierarchy of methods is by far not preserved for the singlet-triplet splitting (Table 3, Figure 5). The functionals BMK, ωB97X, and M06-2X are among the worst options, while B3LYP and B971 perform very well. The T 1 -S 0 gap is in general satisfactorily predicted, for the majority of functionals the error being less than 10%, i.e., smaller than in singlet-singlet excitation energy calculations. This is not surprising, considering that functionals are mostly designed for predicting the energy of the lowest state of a given multiplicity. Noteworthy, M06-2X performs poorly once again, even though it was shown to give excellent results for many aromatic conjugated systems [59]. The good performance of B3LYP is in line with the findings reported by Nguyen et al. [60] who predicted the energies of excited triplet states. Present results for the triplets should be treated with caution, however. The energies and geometric parameters meet the required convergence criteria, but the singlet-triplet splitting is not monotonically decreasing with the size of a carbon chain. We attribute that "noise" to the instability of the obtained triplet-state electronic wavefunction. This shows that even the simplest excited-state calculations do not work like a "black box". An appropriate multi-reference method (CASPT2 type) would likely remedy such instability problems, but the cost, for large systems, can be prohibitive. Table 3. Mean percent errors (with respect to the measured Ar-matrix values) of the singlet-triplet excitation energy calculations performed with various functional/basis set combinations. See Table S3 for ∆ st , and ∆ abs (defined by Equations (3) and (4)). considering that functionals are mostly designed for predicting the energy of the lowe state of a given multiplicity. Noteworthy, M06-2X performs poorly once again, eve though it was shown to give excellent results for many aromatic conjugated systems [59 The good performance of B3LYP is in line with the findings reported by Nguyen et al. [6 who predicted the energies of excited triplet states. Present results for the triplets shoul be treated with caution, however. The energies and geometric parameters meet th required convergence criteria, but the singlet-triplet splitting is not monotonical decreasing with the size of a carbon chain. We attribute that "noise" to the instability o the obtained triplet-state electronic wavefunction. This shows that even the simple excited-state calculations do not work like a "black box". An appropriate multi-referenc method (CASPT2 type) would likely remedy such instability problems, but the cost, fo large systems, can be prohibitive.  (3) and (4)).

Vibrational Frequencies
For the comparison of theoretically derived harmonic vibrational frequencies ω j with experimental fundamental frequencies ν j , a linear relationship was assumed: The least squares fitting procedure supplies the mean value of f : for a set of M normal modes.
The following two statistical parameters were calculated: root mean square error - The f , s, and RMS values have been obtained for the whole training set. Separately, analysis of the bending modes only (for the linear structures) supplied f b , s b , and RMS b .
The results are collected in Tables 4 and 5 and in Supplementary Materials (Tables S116-S119). Differences between the ground-and excited-state scaling factors are generally small and are superimposed on the uncertainty of the determined f coefficients. However, as illustrated by Figure 6, scaling factors are slightly (on the order of 1%) higher in the excited than in the ground states for almost all functionals studied (irrespective of whether the reference data came from the gas phase or from the Kr matrix measurements). We are cautious about the generalization of that finding. The present database of vibrational frequencies is small, and it rarely includes both ground and excited state frequencies for a given vibrational mode. Additionally, the uncertainties inherent (due to a limited spectral resolution) to the experimental determinations are often larger for the excited states than for the ground states and may even approach the RMS errors. Finally, only one specific class of molecules has been tested here, and, to the best of our knowledge, there exist no published reports devoted to the scaling of excited-state harmonic vibrational frequencies.
We can nevertheless confirm the conclusion formulated by Furche and Ahlrichs [14] that the qualities of ground and excited state vibrational frequency computations are similar. It should also be remarked that the overall performance of functionals applied here to ground-state vibrational spectroscopy predictions is not significantly different from what was previously reported [12,13]. n/a n/a n/a n/a n/a n/a BMK 0.974 0.015 26 n/a n/a n/a 0. (a) (b) Figure 6. Correlation between the vibrational frequency scaling factors obtained for the ground electronic states, fg, and for the excited states, fe. Orange line represents the best fit of the fe = slope × fg function. Yellow strap marks the 3σ confidence interval (the green line corresponding to slope = 1 falls out of 3σ). Computations were performed with all functionals listed in Tables 4 and 5 and with the aug-cc-pVTZ basis set. (a) Reference data taken from gas-phase measurements; the uncertainty of slope value is 0.0019. (b) Reference data taken from Kr-matrix measurements; the uncertainty of slope value is 0.004.

Recommendations
Presently collected data and the published results unable us to formulate certain recommendations for the TD-DFT calculations on cyanopolyyne-family molecules. These can likely be extended also to other polyynes. Our study shows, however, that it is necessary to test a given TD-DFT method for a select class of molecules and electronic states, without relying too heavily on benchmark results reported for other species.
While it is not possible to indicate a functional perfectly suited for the description of both the ground and the lowest excited states of the tested molecules, the use of ωB97X and CAM-B3LYP can be advised when the goal consists in predicting the singlet-singlet excitation energies. The above two functionals, however, are not the best choice when it comes to the computation of triplet-singlet spacings or the ground-state vibrational frequencies. B3LYP or B971 should rather be applied for these tasks, but then the description of excited singlet states would suffer. The B972 functional seems to produce the most balanced results. It works decently well in predicting the excitation energies to singlet states, the excited singlet state vibrational frequencies, singlet-triplet splittings, and the ground-state vibrational spectroscopy. The appropriate choice of a basis set is either aug-cc-pVTZ or def2TZVPD. Figure S1 illustrates how the calculation method may affect the predicted band positions and the associated oscillator strengths. It should be remarked, however, that neither the inclusion of important effects coming from the vibronic coupling nor predictions of the vibrational structure of electronic transitions were within the scope of this work.

Electronic Excitation Energy of Longer Cyanoacetylenes.
The S1-S0, S2-S0, and T1-S0 transitions are adequately described in terms of HOMO-LUMO excitations. The HOMO and LUMO orbitals, π and π*, respectively (Figures 7 and  8), do not exhibit any helical allene-like [61,62] arrangement. Electron density maxima in HOMOs correspond to the triple bond positions. The shape of LUMOs is unusual. At the Figure 6. Correlation between the vibrational frequency scaling factors obtained for the ground electronic states, f g , and for the excited states, f e . Orange line represents the best fit of the f e = slope × f g function. Yellow strap marks the 3σ confidence interval (the green line corresponding to slope = 1 falls out of 3σ). Computations were performed with all functionals listed in Tables 4 and 5 and with the aug-cc-pVTZ basis set. (a) Reference data taken from gas-phase measurements; the uncertainty of slope value is 0.0019. (b) Reference data taken from Kr-matrix measurements; the uncertainty of slope value is 0.004.

Recommendations
Presently collected data and the published results unable us to formulate certain recommendations for the TD-DFT calculations on cyanopolyyne-family molecules. These can likely be extended also to other polyynes. Our study shows, however, that it is necessary to test a given TD-DFT method for a select class of molecules and electronic states, without relying too heavily on benchmark results reported for other species.
While it is not possible to indicate a functional perfectly suited for the description of both the ground and the lowest excited states of the tested molecules, the use of ωB97X and CAM-B3LYP can be advised when the goal consists in predicting the singlet-singlet excitation energies. The above two functionals, however, are not the best choice when it comes to the computation of triplet-singlet spacings or the ground-state vibrational frequencies. B3LYP or B971 should rather be applied for these tasks, but then the description of excited singlet states would suffer. The B972 functional seems to produce the most balanced results. It works decently well in predicting the excitation energies to singlet states, the excited singlet state vibrational frequencies, singlet-triplet splittings, and the ground-state vibrational spectroscopy. The appropriate choice of a basis set is either augcc-pVTZ or def2TZVPD. Figure S1 illustrates how the calculation method may affect the predicted band positions and the associated oscillator strengths. It should be remarked, however, that neither the inclusion of important effects coming from the vibronic coupling nor predictions of the vibrational structure of electronic transitions were within the scope of this work.

Electronic Excitation Energy of Longer Cyanoacetylenes
The S 1 -S 0 , S 2 -S 0 , and T 1 -S 0 transitions are adequately described in terms of HOMO-LUMO excitations. The HOMO and LUMO orbitals, π and π*, respectively (Figures 7 and 8), do not exhibit any helical allene-like [61,62] arrangement. Electron density maxima in HO-MOs correspond to the triple bond positions. The shape of LUMOs is unusual. At the molecular extremities LUMO resembles typical π-antibonding orbitals; inside the chain, however, it behaves rather like a π-bonding, with electron density maxima coinciding with the locations of single bonds. This indicates that the HOMO-LUMO excitations cause a shift from the polyyne-like structure towards a cumulenic one. Such a conjecture is confirmed with the observed bond length alternation (BLA; defined as the difference in length of the neighboring bonds near the center of the molecule [21]). For the ground state, as can be seen in Figure 9, even the longest chains show substantial BLA. After the electronic excitation, BLA drops down drastically (a similar theoretical result was reported for hexayne [32]). In T 1 , it oscillates around 0.003 Å, and in the S 1 and S 2 states it is around 0.01 Å. Such low values are characteristic for cumulenes and allenes [63].
Molecules 2022, 27, x FOR PEER REVIEW 12 of 19 molecular extremities LUMO resembles typical π-antibonding orbitals; inside the chain, however, it behaves rather like a π-bonding, with electron density maxima coinciding with the locations of single bonds. This indicates that the HOMO-LUMO excitations cause a shift from the polyyne-like structure towards a cumulenic one. Such a conjecture is confirmed with the observed bond length alternation (BLA; defined as the difference in length of the neighboring bonds near the center of the molecule [21]). For the ground state, as can be seen in Figure 9, even the longest chains show substantial BLA. After the electronic excitation, BLA drops down drastically (a similar theoretical result was reported for hexayne [32]). In T1, it oscillates around 0.003 Å, and in the S1 and S2 states it is around 0.01 Å. Such low values are characteristic for cumulenes and allenes [63].   molecular extremities LUMO resembles typical π-antibonding orbitals; inside the chain, however, it behaves rather like a π-bonding, with electron density maxima coinciding with the locations of single bonds. This indicates that the HOMO-LUMO excitations cause a shift from the polyyne-like structure towards a cumulenic one. Such a conjecture is confirmed with the observed bond length alternation (BLA; defined as the difference in length of the neighboring bonds near the center of the molecule [21]). For the ground state, as can be seen in Figure 9, even the longest chains show substantial BLA. After the electronic excitation, BLA drops down drastically (a similar theoretical result was reported for hexayne [32]). In T1, it oscillates around 0.003 Å, and in the S1 and S2 states it is around 0.01 Å. Such low values are characteristic for cumulenes and allenes [63].    [64]. Experimental values (with 1σ experimental uncertainty) are based on microwave spectroscopic measurements of the ground vibrational states [65].
B972/aug-cc-pVTZ computations lead to ground-state BLA in the vicinity of 0.1 Å. Similar predictions were reported for variously capped polyynes [63,66]. The B3LYP result for a chain oligomer C72 is 0.088 Å [64], i.e., close to 0.095 Å obtained here for C30N2. The recommended BLA of a long polyyne chain [64] falls between the values currently predicted by B972 and ωB97X for the chains of the length 3-4 nm. Experimental bond lengths, available for HC5N, HC7N, HC9N, HC11N [65], also lead to the BLA confined between the B972-and ωB97X-derived values, usually closer to B972. As can be seen in Figure 9, the experimental BLA does not necessarily change monotonically with the size of the carbon chain, unlike the predictions for S0. However, these latter refer to the equilibrium bond lengths, i.e., the ones at the minimum of the potential energy surface. The experiment, on the other hand, supplies vibrationally averaged values for the ground vibrational state. That difference may alter the BLA. Nevertheless, the disagreement between theory and experiment is only a bit higher than the typical accuracy of BLA predictions accomplished with the DFT methods [67].
The transition from polyyne to cumulene structure is visible in the vibrational spectroscopy as well. Frequencies (Tables S117-S168) assigned to the triple bond stretching modes decrease after the electronic excitation. An opposite comportment is observed for the single bond stretches.
Significant changes in the electronic structure produced upon excitation may be of practical importance. A polyyne molecule can be classified, in its ground state, as a semiconductor. Singlet-triplet separation is 1-2 eV for the chain lengths of 4-1.5 nm, and the electrical conductivity likely decreases for longer molecules [68]. The cumulene chains, on the other hand, behave as electrical conductors and that property increases with their Figure 9. Bond length alternation plotted as a function of the number of carbon atoms in HC 2n+1 N (a) and C 2n+2 N 2 (b) families. The upper horizontal scale gives approximate lengths of the carbonnitrogen chains. The black dash horizontal line represents the BLA value estimated for a long polyyne [64]. Experimental values (with 1σ experimental uncertainty) are based on microwave spectroscopic measurements of the ground vibrational states [65].
B972/aug-cc-pVTZ computations lead to ground-state BLA in the vicinity of 0.1 Å. Similar predictions were reported for variously capped polyynes [63,66]. The B3LYP result for a chain oligomer C 72 is 0.088 Å [64], i.e., close to 0.095 Å obtained here for C 30 N 2 . The recommended BLA of a long polyyne chain [64] falls between the values currently predicted by B972 and ωB97X for the chains of the length 3-4 nm. Experimental bond lengths, available for HC 5 N, HC 7 N, HC 9 N, HC 11 N [65], also lead to the BLA confined between the B972-and ωB97X-derived values, usually closer to B972. As can be seen in Figure 9, the experimental BLA does not necessarily change monotonically with the size of the carbon chain, unlike the predictions for S 0 . However, these latter refer to the equilibrium bond lengths, i.e., the ones at the minimum of the potential energy surface. The experiment, on the other hand, supplies vibrationally averaged values for the ground vibrational state. That difference may alter the BLA. Nevertheless, the disagreement between theory and experiment is only a bit higher than the typical accuracy of BLA predictions accomplished with the DFT methods [67].
The transition from polyyne to cumulene structure is visible in the vibrational spectroscopy as well. Frequencies (Tables S117-S168) assigned to the triple bond stretching modes decrease after the electronic excitation. An opposite comportment is observed for the single bond stretches.
Significant changes in the electronic structure produced upon excitation may be of practical importance. A polyyne molecule can be classified, in its ground state, as a semiconductor. Singlet-triplet separation is 1-2 eV for the chain lengths of 4-1.5 nm, and the electrical conductivity likely decreases for longer molecules [68]. The cumulene chains, on the other hand, behave as electrical conductors and that property increases with their length [68]. Using light to toggle between the two structures is a challenge. So far, it has been demonstrated, to a limited extent, for a C 12 polyyne capped with the adamantyl groups [69]. Figure 2 illustrates a decline in electronic excitation energies with the increasing molecular size, well known from the other studies on polyynes [18,56,70,71].

Materials and Methods
Gaussian 09 Rev. B.01 [72] was applied for benchmark computations and Gaussian 16 Rev. B.01 [73] for studying the longest CPYs. Singlet-state excitation energies were calculated using the standard TD-DFT approach [14,[74][75][76][77]. The S 0 -T 1 energy gaps were obtained as ∆SCF, i.e., as the difference in energies of the lowest triplet and singlet states. The triplet energy was given by unrestricted DFT computations (α and β orbitals relaxed independently). The benchmark studies used the XQC option, which automatically switched from the default DIIS-type [78,79] to a quadratically convergent [80] SCF technique in case of a numerical problem. For longer molecules, where the XQC procedure did not converge, the YQC algorithm was employed. In all calculations (i) two-electron integrals were computed with pruned UltraFine grid and (ii) tight convergences criteria were applied for geometry optimization and electronic energy computations.
Computations were carried out at our local computer cluster equipped with Intel Xeon CPU E5-1660 v4 and Intel Xeon CPU E3-1240 processors.

Conclusions
The performance of DFT functionals and basis sets has been benchmarked with respect to electronic spectroscopy predictions for cyanopolyynes. While we cannot indicate a single functional ideally suited for describing both the ground and the lowest excited electronic states, B972 generally provided reasonably balanced results. The ωB97X and CAM-B3LYP functionals are, however, a better choice for the derivation of singlet-singlet excitation energies, whereas B3LYP and B971 should rather be applied for predicting the triplet-singlet spacing values or the ground-state vibrational frequencies. The aug-cc-pVTZ and def2TZVPD basis sets offer a good compromise between the cost and quality of computations.
Present results provide the functional-and basis set-dependent scaling factors for the correction of vibrational frequencies obtained with the harmonic approximation. Our recommendation is to use the same scaling factors for the ground and for the excited electronic states.
This study shows that DFT calculations should be preceded by appropriate benchmarking with regard to the class of molecules and electronic states under study. Even then, the results are subject to considerable uncertainty.
Our predictions pertaining to relatively long cyanopolyynes indicate that their optical excitation leads to a shift from the polyyne-to cumulene-like electronic structures. These latter structures conduct electricity, which is of possible interest for the design of future nanowires. We expect the first electronic states of 1.5-4 nm long molecules (C 12 N 2 , C 14 N 2 , C 16 N 2 , C 18 N 2 , C 20 N 2 , C 22 N 2 , C 24 N 2 , C 26 N 2 , C 30 N 2 , HC 13 N, HC 15 N, HC 19 N, HC 21 N, HC 23 N, HC 25 N, HC 27 N) to be excited with green-orange light, while phosphorescence may occur in the red or near-infrared range.