Conformers of dehydrogenated glycine isomers

We report a comprehensive ab initio investigation of the conformers of dehydrogenated glycine radicals using the STO‐3G, 3‐21G, and aug‐cc‐pVDZ (aVDZ) basis sets and the UHF and UMP2 (H2N‐CH‐COOH and HN‐CH2‐COOH) as well as MCSCF and MRCI (H2N‐CH2‐COO) methods via two different conformational search strategies generating initial structures for optimizations by (a) removing H atoms from glycine conformers and (b) scanning torsional angles describing internal rotation along the CC, CN, and CO (except for H2N‐CH2‐COO) bonds of the radicals. We find four H2N‐CH‐COOH {InCH, IInCH, IIInCH, IVnCH} and seven HN‐CH2‐COOH {IpNH, IIpNH, IIInNH, IVpNH VnNH, VIpNH, VIIpNH} conformers with classical(adiabatic) relative energies of {0.00(0.00), 1.57(1.55), 5.25(5.03), 9.85(9.72)} and {0.00(0.00), 0.78(1.06), 1.93(2.08), 3.34(3.16), 3.39(3.29), 5.00(4.86), 9.27(8.87)} kcal/mol, respectively, obtained with UCCSD(T)‐F12b/aug‐cc‐pVTZ(+UCCSD(T)‐F12b/aVDZ ZPE correction) and four H2N‐CH2‐COO {IpCOO, IInCOO, IIIpCOO, IVnCOO} conformers with MRCI‐F12+Q/aVDZ(+MRCI/aVDZ ZPE correction) energies of {0.00(0.00), 1.65(1.64), 1.78(1.75), 2.21(2.21)} kcal/mol, where n and p denote C1 and Cs symmetry. The MRCI‐F12+Q[UCCSD(T)‐F12b] InCH → IpNH and InCH → IpCOO classical(adiabatic) isomerization energies are 18.51(17.32)[21.20(20.01)] and 31.88(31.66) kcal/mol, respectively.

few studies [13,[16][17][18]21] considered the lowest energy conformer of HN-CH 2 -COOH and others [13,18] reported several conformers of all the three isomer radicals derived from homolytic CH, NH, and OH bond dissociation. The above-mentioned theoretical studies employed density functional theory (usually B3LYP) and the MP2 method with double-and triple-zeta basis sets (usually 6-31G* and 6-311+ +G**). [13][14][15][16][17][18][19][21][22][23] The use of the more sophisticated and accurate CCSD(T) method is rare and only employed for single-point energy computations. [14,21] In the present study, we report a comprehensive ab initio study of the conformers of gas-phase glycine radicals considering CH, NH, and OH bond dissociations. We go beyond the accuracy of the previous studies by applying the explicitly correlated CCSD(T)-F12b method [25] to obtain benchmark structures and relative energies of the dehydrogenated glycine isomers. Besides the high-level ab initio investigation of the glycine radicals, we test different computational strategies to determine all the possible conformers of the title system, which techniques may become useful for mapping the conformational space of other similar or even larger systems.
The motivation of the present study is multiple-fold: (a) We determine conformers of glycine with the state-of-the-art explicitly correlated CCSD(T)-F12b method, for the first time, thereby confirming and/or improving previous work. [1,4] (b) We aim to find new conformers for the dehydrogenated glycine radicals, thereby complementing previous incomplete studies. [13][14][15][16][17][18][19][20][21][22][23][24] (c) We provide benchmark structures and energetics for the conformers of glycine radicals. (d) As the topology of the potential energy surface may sensitively depend on the level of electronic structure theory, we test different ab initio methods, including also multi-reference techniques, and basis sets to find radical conformers. Knowing the performance of the low-level methods can provide useful guidance for the investigations of larger systems, when the use of high-level theories is not feasible. (e) The present benchmark characterization of the dehydrogenated glycine radicals may be the first step toward the study of the OH + glycine reaction.
In Section 2, we describe the computational details including the introduction of two different conformational search strategies. The results are presented and discussed in Section 3. The article ends with summary and conclusions in Section 4.

| Conformers of glycine
Following the pioneering electronic structure studies reporting eight conformers (see Figure 1) of the gaseous glycine, [1,4] we optimize these conformers and compute their harmonic frequencies using the second-order Møller-Plesset perturbation theory (MP2) [26] combined with the correlation-consistent aug-cc-pVDZ basis set. [27] Then, we further optimize the obtained structures by the explicitly correlated coupled-cluster singles, doubles, and perturbative triples method (CCSD(T)-F12b) [25] using the aug-cc-pVDZ and aug-cc-pVTZ basis sets. [27] Besides MP2, the harmonic frequencies are also computed at the CCSD(T)-F12b/aug-cc-pVDZ level of theory.

| Conformers of dehydrogenated glycine isomers
Our goal is to determine all the possible conformers of the dehydrogenated glycine radicals. To achieve this goal we investigate two different strategies and various levels of electronic structure theory as detailed below.

| Strategy I
As a first, simple, chemically motivated strategy we remove one H atom from the central C atom, the amino or the carboxyl group of each of the eight glycine conformers. Since the two CH 2 and NH 2 hydrogen atoms are equivalent in the three C s glycine geometries and different in the five C 1 structures, the H abstraction leads to 3 + 2 × 5 = 13, 3 + 2 × 5 = 13, and eight different initial radical geometries for the H 2 N-CH-COOH, HN-CH 2 -COOH, and H 2 N-CH 2 -COO isomers, respectively. In the case of H 2 N-CH-COOH and HN-CH 2 -COOH we perform geometry optimizations starting from the above two times 13 initial structures using the following levels of theory: UHF/STO-3G, UHF/3-21G, UHF/aug-cc-pVDZ, UMP2/STO-3G, UMP2/3-21G, and UMP2/aug-cc-pVDZ. [28][29][30][31]27] For the 8 H 2 N-CH 2 -COO geometries we experience convergence problems in Hartree-Fock [32] (both ROHF and UHF); thus, the use of the multiconfigurational self-consistent field (MCSCF) [33] and multi-reference configuration interaction (MRCI) [34] methods is found to be necessary. Therefore, we optimize the H 2 N-CH 2 -COO geometries using the MCSCF/STO-3G, MCSCF/3-21G, MCSCF/aug-cc-pVDZ, MRCI/STO-3G, MRCI/3-21G, and MRCI/aug-cc-pVDZ levels of theory starting from the eight initial structures. The MCSCF computations utilize a small active space of five electrons on three spatial orbitals and we compute the ground electronic state only. During all the correlation computations in this study, the core electrons are kept frozen.

| Strategy II
In order to ensure that we have found all the possible conformers with Strategy I, we perform a more systematic mapping of the conformational space of the glycine radicals. Starting from the lowest-energy conformer obtained by Strategy I for each of the three isomers, we generate 6 3 = 216, 6 3 = 216, and 6 2 = 36 initial geometries by varying 3, 3, and 2 torsion angles between 0 and 360 with 60 steps (

| Benchmark structures and energies
The final UMP2/aug-cc-pVDZ conformers obtained by Strategies I and II are further optimized using the ROHF-based UCCSD(T)-F12b [35] method with the aug-cc-pVDZ (geometry and frequency computation) and aug-cc-pVTZ (geometry) basis sets. Thus, the best classical relative energies of the H 2 N-CH-COOH and HN-CH 2 -COOH conformers are obtained at the UCCSD(T)-F12b/aug-cc-pVTZ level of theory and the adiabatic relative energies include UCCSD(T)-F12b/ aug-cc-pVDZ zero-point energy corrections. For the H 2 N-CH 2 -COO conformers the classical relative energies are determined using the Davidson-corrected [36] MRCI+Q [34] and MRCI-F12+Q [37] methods with the aug-cc-pVDZ basis set at the MRCI/aug-cc-pVDZ geometries and the adiabatic relative energies are obtained utilizing MRCI/augcc-pVDZ zero-point energy (ZPE) corrections. In order to compare the energies of all the glycine radical isomers, MRCI+Q/aug-cc-pVDZ and MRCI-F12+Q/aug-cc-pVDZ energies are also computed for the H 2 N-CH-COOH and HN-CH 2 -COOH conformers at the UCCSD(T)-F12b/ aug-cc-pVTZ geometries. Furthermore, MRCI/aug-cc-pVDZ frequencies are determined for the lowest-energy H 2 N-CH-COOH conformer to obtain ZPE-corrected isomerization energy between H 2 N-CH-COOH and H 2 N-CH 2 -COO. All the ab initio computations in this study are performed using the MOLPRO [38] program package.

| Conformers of glycine
The structures and relative energies of the eight conformers (minima) of glycine are given in Figure 1 and Table 1, respectively. The notation of the conformers follows previous studies, [1][2][3][4] that is, roman numbers increase with the energy of C s structures and p and n refer to planar (C s symmetry) and nonplanar (C 1 symmetry) arrangements of the N-C-COOH atoms, respectively. Three conformers (Ip, VIp, and VIIp) have C s symmetry, whereas in five cases (IIn, IIIn, IVn, Vn, and VIIIn) lone-electron-pair repulsion effects favor symmetry breaking; thus, the minima are nonsymmetric and the C s structures (not shown in Figure 1) correspond to saddle points. [1] The symmetry-breaking stabilizer effects change the energy order of the III and IV conformers, that is, IVn is below IIIn by about 0.5 kcal/mol in agreement with previous studies. [1,4] For the glycine conformers the MP2/aug-cc-pVDZ level of theory provides remarkably accurate relative energies with only about 0.1 kcal/mol mean and 0.2 kcal/mol maximum differences from the CCSD(T)-F12b/aug-cc-pVDZ results as shown in Table 1. The CCSD (T)-F12b relative energies obtained with the aug-cc-pVDZ and augcc-pVTZ basis sets agree within 0.00-0.03 kcal/mol showing the excellent basis-convergence of the explicitly-correlated CCSD(T)-F12b method. The present CCSD(T)-F12b/aug-cc-pVTZ classical relative energies are in excellent agreement (the average absolute deviation is less than 0.1 kcal/mol) with the final predictions of Császár, [1] confirming that the "conservative" error bar estimate of ±0.3 kcal/mol given in Császár [1] was really conservative. Furthermore, the present benchmark classical relative energies reproduce the CCSD(T)/complete-basis-set(CBS) results of Balabin [4]  confirming the Ip, IIn, IVn, IIIn, Vn, VIp, VIIp, and VIIIn energy order of previous theoretical predictions. [1,4]  COO radicals as shown in Figure 3. Similar to glycine, the radical conformers are denoted by roman numbers reflecting the UMP2/aug-cc-pVDZ or MRCI/aug-cc-pVDZ energy order, p and n denote C s and C 1 point-group symmetry, respectively, and subscripts CH, NH, and COO denote the radical isomer.
All the four H 2 N-CH-COOH conformers are nonsymmetric and related to each other by internal rotations along the CC and CO axes. The assignment of the conformers (see Figure 3), obtained by UMP2/aug-cc-pVDZ computations initiated from the lower-level optimized structure, are shown in parentheses. b Initial structure of the geometry optimization obtained by removing the indicated H atom from the given glycine conformer (see Figure 1). c UHF convergence problems or the optimization did not converge within 100 steps. d Imaginary frequencies are obtained.

T A B L E 3
Relative energies (kcal/mol) of HN-CH 2 -COOH conformers obtained from the eight glycine conformers at different levels of theory a The assignment of the conformers (see Figure 3), obtained by UMP2/aug-cc-pVDZ computations initiated from the lower-level optimized structure, are shown in parentheses. b Initial structure of the geometry optimization obtained by removing the indicated H atom from the given glycine conformer (see Figure 1). In the case of the HN-CH 2 -COOH radical, all the optimizations converge successfully (see Table 3). The UMP2/aug-cc-pVDZ level of theory results in 7 conformers, from which only two (IIIn NH and Vn NH ) do not have C s symmetry as shown in Figure 3. VIIIn- The assignment of the conformers (see Figure 3), obtained by MRCI/aug-cc-pVDZ computations initiated from the lower-level optimized structure, are shown in parentheses. b Initial structure of the geometry optimization obtained by removing the indicated H atom from the given glycine conformer (see Figure 1). Additional conformers (minima) are not found using the lower levels of theory.
For the H 2 N-CH 2 -COO radical MRCI/aug-cc-pVDZ finds four conformers as shown in Table 4 and Figure 3. There are two C s conformers (Ip COO and IIIp COO ), which differ in the conformation of the COO group, and there are two C 1 conformers (IIn COO and IVn COO ), where the NH 2 group is twisted (Figure 3). (Note that an additional fifth conformer is also obtained from IIn, which seems to be a IVn COOlike TS structure with C s symmetry, but its saddle-point character cannot be confirmed, because the frequency computation does not converge for this conformer.) As shown in Table 4 Figure 3) are assigned by UMP2/aug-cc-pVDZ computations initiated from the lower-level optimized structures [Color figure can be viewed at wileyonlinelibrary.com] convergence problem does not occur and IVn COO is obtained from both IIn and Vn, and in some cases from VIIp, the energies of the IVn COO -like conformers are always different (see Table 4). Thus, additional conformers, which are IVn COO -like minima at the lowest energies and C s saddle points or twisted C 1 minimum (MCSCF/STO-3G) at the higher energies (highest energy MRCI/3-21G conformer has no symmetry with a twisted COO group), are found at lower levels of theory, which all result in the same IVn COO conformer with further MRCI/aug-cc-pVDZ optimizations.   Table 3). The UMP2 method with the 3-21G and STO-3G basis sets basically provide the same seven conformers as in the case of the aug-cc-pVDZ basis. It is important to note that Strategy II finds VIIp NH 17 and 5 times with 3-21G and STO-3G, respectively; whereas this conformer was missing using the above small basis sets with Strategy I. Using the UHF method the conformational potential is more rugged resulting in many conformers; some of them correspond to the ones found with Strategy I and the other usually have small probabilities as shown in Figure 6. These small probability conformers vanish with further optimizations at the UMP2/aug-cc-pVDZ level and at the end all the conformers relax to the structures shown in Figures 3 and 4. .52 kcal/mol both converge to the same IVn COO conformer at the MRCI/aug-cc-pVDZ level. Interestingly, Strategy I found a sixth conformer at 3.43 kcal/mol, which was also assigned to IVn COO ; however, F I G U R E 7 Number of the different initial geometries resulted in the same conformer with relative energy (E rel ) within 0.01 kcal/mol of H 2 N-CH 2 -COO obtained from 36 initial structures at different levels of theory. The MRCI conformers (for structures see Figure 3) are assigned by MRCI/aug-cc-pVDZ computations initiated from the lower-level optimized structures. The MCSCF conformers are not assigned [Color figure can be viewed at wileyonlinelibrary.com] this sixth conformer is not seen with Strategy II. Using MRCI/STO-3G six conformers are obtained, which all assigned to the same four minima with further MRCI/aug-cc-pVDZ optimizations as shown in Figure 7.

| Strategy II
Strategy I gave four conformers at the MRCI/STO-3G level, but IIIp COO was missing, which is found with Strategy II, albeit only once. Furthermore, Strategy II finds an additional conformer at 2.05 kcal/mol, which is assigned to IVn COO . In the case of the MCSCF method 14, 14, and 15 conformers are found with the STO-3G, 3-21G, and aug-cc-pVDZ basis sets, respectively, showing again that the potential energy surface is more structured at lower levels of theory. Here many conformers are found only 1 or 2 times from the 36 optimizations, and only a few of them, which were also obtained with Strategy I, are found 4-9 times.
Owing to the large number of conformers, we do not perform MRCI/ aug-cc-pVDZ optimization for each geometry, nevertheless, on the basis of our previous findings we assume that all these conformers would result in the same four minima as seen using MRCI (Figure 7).
aug-cc-pVDZ conformers, give four minima for H 2 N-CH-COOH and seven minima (and four transition states) for HN-CH 2 -COOH; thus, none of the UMP2/aug-cc-pVDZ conformers disappears at higher levels of theory. As Table 5 shows the MP2/aug-cc-pVDZ level sig- COOH conformers are given in Table 6 (minima) and Table 7 Table 8. The Davidson corrections (+Q), which estimate the dynamical correlation effects beyond double excitations, are found to be small (0.00-0.04 kcal/mol) as seen in Table 8.
Furthermore, the basis-set effects beyond aug-cc-pVDZ are also negligible, because the standard MRCI+Q and explicitly correlated MRCI-F12+Q relative energies again agree within 0.00-0.04 kcal/ mol ( Table 8).
Comparison of the benchmark classical and adiabatic energies of all the conformers of the H 2 N-CH-COOH, HN-CH 2 -COOH, and H 2 N-CH 2 -COO radicals is shown in

| SUMMARY AND CONCLUSIONS
We have determined the first explicitly-correlated CCSD(T)-F12b/ aug-cc-pVTZ structures and relative energies of the eight conformers of glycine confirming the predictions of previous studies. [1,4] Furthermore, we report a comprehensive ab initio investigation of the conformers and isomers of dehydrogenated glycine radicals revealing 4, 7 (+4 TS), and 4 conformers for H 2 N-CH-COOH, HN-CH 2 -COOH, and H 2 N-CH 2 -COO, respectively. The four conformers of H 2 N-CH-COOH are nonsymmetric (C 1 ), whereas the conformers of the other two isomers have either C 1 or C s point-group symmetry.
We have used two different strategies for the conformation search employing various ab initio methods and basis sets. Strategy I is motivated by chemical intuition suggesting initial geometries by removing different H atoms from the eight known conformers of glycine. Strategy II systematically maps the conformational space of the glycine radicals generating initial structures for geometry optimizations by scanning the most important torsional coordinates with 60 steps as was previously done for the amino acid threonine. [39] The conclusions of the detailed search using the different strategies and ab initio levels can be summarized as follows: 1 Strategy I often finds all the conformers. (VIIp NH is not obtained at UHF/3-21G, UMP2/STO-3G, and UMP2/3-21G; IVn COO is not found at MCSCF/aug-cc-pVDZ; IIIp COO is missing at MRCI/ STO-3G.) 2 The conformational space is usually more structured at the UHF and MCSCF levels than with the UMP2 and MRCI methods. However, the additional conformers found at the lower levels of theory disappear when further optimizations are performed using higherlevel correlation methods.
3 Strategy II with the UMP2 and MRCI methods provides qualitatively the same conformers with the 3-21G and aug-cc-pVDZ basis sets, whereas the small STO-3G basis predicts additional conformers, which converge to the known minima with larger basis sets. 4 Both Strategies I and II give the same conformers at UMP2/aug-cc-pVDZ (H 2 N-CH-COOH and HN-CH 2 -COOH) and MRCI/aug-cc-pVDZ (H 2 N-CH 2 -COO) levels of theory. 5 All the conformers found at the UMP2/aug-cc-pVDZ level of theory can be confirmed using the CCSD(T)-F12b method with the aug-cc-pVDZ and aug-cc-pVTZ basis sets. 6 For larger systems we recommend the initial conformational search at a computationally cheap low-level of theory using either Strategy I or II, followed by higher-level optimizations where several conformers are likely to disappear. In the case of Strategy II the MP2 method with the 3-21G basis is an economic choice, for the initial search there is no need for the larger aug-cc-pVDZ basis.
The present study focuses on ab initio methods; however, it is important to note that density functional theory may also become useful for conformational searches of larger systems. For example, a recent study found excellent performance of the B3LYP-D3BJ and ωB97X-V functionals for the determination of conformational energies of amino acids with N-and C-termination. [40] Benchmark static electron correlation is not significant for the H 2 N-CH-COOH and HN-CH 2 -COOH conformers as the T 1 -diagnostic [41] values are only around 0.017. For the H 2 N-CH 2 -COO geometries, we have found convergence problems when using single-configuration methods; thus, in this case only multi-reference data could be obtained.
Besides the new benchmark properties and the first comprehensive characterization of the dehydrogenated glycine conformers, the present ab initio investigation opens the door for more detailed studies of the H-abstraction reactions of free radicals with glycine. Future work may consider solvation effects to mimic biological environment.
Furthermore, the conformational search strategies could be used to find conformers of similar or even larger complex molecular systems.