Approximate Spin Projection for Broken-Symmetry Method and Its Application

A broken-(spin) symmetry (BS) method is now widely used for systems that involve (quasi) degenerated frontier orbitals because of their lower cost of computation. The BS method splits up-spin and down-spin electrons into two different special orbitals, so that a singlet spin state of the degenerate system is expressed as a singlet biradical. In the BS solution, therefore, the spin symmetry is no longer retained. Due to such spin-symmetry breaking, the BS method often suffers from a serious problem called a spin contamination error, so that one must eliminate the error by some kind of projection method. An approximate spin projection (AP) method, which is one of the spin projection procedures, can eliminate the error from the BS solutions by assuming the Heisenberg model and can recover the spin symmetry. In this chapter, we illustrate a theoretical background of the BS and AP methods, followed by some examples of their applications, especially for calculations of the exchange interaction and for the geometry optimizations.

Physical properties of molecules are sometimes discussed by using several parameters such as an exchange integrals (J ab ), on-site Coulomb repulsion, and transfer integrals of Heisenberg and Hubbard Hamiltonians, respectively, in material physics [31][32][33][34][35]. In recent years, on the other hand, direct predictions of such electronic structures, molecular structure, and physical properties of those metal complexes are fairly realized by the recent progress in computers and computational methods. In this sense, theoretical calculations are now one of the powerful tools for understanding of such systems. However, those systems are, in a sense, still challenging subjects because they are usually large and orbitally degenerated systems with localized electron spins (localized orbitals). The localized spins are caused by an electron correlation effect called a static (or a non-dynamical) correlation [36]. In addition, a dynamical correlation effect of core electrons also must be treated together with the static correlation in the case of the metal complexes. A treatment of both the static correlation and the dynamical correlation in large molecules is still a difficult task and a serious problem in this field. For those systems, a standard method for the static and dynamical correlation corrections is a complete active space (CAS) method [37][38] or a multi-reference (MR) method [39] that considers all configuration interaction in active valence orbitals, together with the second-order perturbation correction, e.g., CASPT2 or MPMP2 methods. In addition to these methods, recently, other multiconfiguration methods such as DDCI [40][41][42], CASDFT [43][44][45], MRCC [46][47][48], and DMRG-CT [49][50][51] methods are also proposed for the same purpose. These newer methods are developing and seem to be promising tools in terms of accuracy; however, real molecules such as polynuclear metal complexes are still too large to treat computationally with those methods at this state. An alternative way is a broken-symmetry (BS) method, which approximates the static correlation with a lower cost of computation [52][53][54][55]. The BS method (or commonly known as an unrestricted (U) method) splits up and down spins (electrons) into two different spatial orbitals (it is sometimes called as different orbitals for different spins; DODS), so a singlet spin state of the orbitally degenerated system is expressed as a singlet biradical, namely, the BS singlet [55]. The BS method such as the unrestricted Hartree-Fock (UHF) and the unrestricted DFT (UDFT) methods are now widely used for the first principle calculations of such large degenerate systems. In this sense, the BS method seems to be the most possible quantum chemical approach for the polynuclear metal complexes, although it has a serious problem called the spin contamination error [56][57][58][59][60][61][62][63][64][65]. Therefore one must eliminate the error by some kind of projection method. An approximate spin projection (AP) method, which is one of the spin projection procedures, can eliminate the error from the BS solutions and can recover the spin symmetry. In this chapter, we illustrate a theoretical background of the BS and AP methods, followed by some examples of their applications.

Theoretical background of AP method
In this section, the theoretical background of the BS and AP methods for the biradical systems is explained with the simplest two-spin model (e.g., a dissociated H 2 ) as illustrated in Figure 1(a).

Broken-symmetry (BS) solution and approximate spin projection (AP) methods for the (two-spin) biradical state
In the BS method, the spin-polarized orbitals are obtained from HOMO-LUMO mixing [55][56]. For example, HOMO orbitals for up-spin (ψ HOMO ) and down-spin (ψ HOMO ) electrons of the simple H 2 molecule are expressed as follows (Figure 1 where 0 ≤ θ ≤ 45 and ψ HOMO and ψ LUMO express HOMO and LUMO orbitals of spin-adapted (SA) (or spin-restricted (R)) calculations, respectively, as illustrated in Figure 1 (b). And the wavefunction of the BS singlet (e.g., unrestricted Hartree-Fock (UHF)) becomes À b Eq. (11) is the so-called Yamaguchi equation to calculate J ab values with the AP procedure, which is simply denoted by J ab here. The calculated J ab value can explain an interaction between two spins. If a sign of calculated J ab value is positive, the HS, i.e., ferromagnetic coupling state, is stable, while if it is negative, the LS, i.e., antiferromagnetic coupling state is stable. Therefore, one can discuss the magnetic interactions in a given system.

Approximate spin projection for BS energy and energy derivatives
Because J ab calculated by Eq. (11) is a value that the spin contamination error is approximately eliminated, it should be equal to J ab value calculated by the approximately spin-projected LS energy (E LS AP ) as Here, we assume b where (14) and Then, we explain about derivatives of this spin-projected energy (E LS AP ). In order to carry out the geometry optimization using the AP method, an energy gradient of E LS AP is necessary. E LS AP can be expanded by using Taylor expansion: where G LS AP R ð Þ and F LS AP R ð Þ are the first and second derivatives (i.e., gradient and Hessian) of [62][63][64][65]; R LS AP and R are a stationary point of E LS AP R ð Þ and a present position, respectively; and X is a position vector ( where G LS BS and G HS are the first energy derivatives (energy gradients) of the BS and the HS states, respectively. As mentioned above, the spin contamination in the HS state is negligible, so that b S 2 D EHS is usually a constant. Then ∂α R ð Þ=∂R can be written as By using Eqs. (16) and (17), the AP optimization can be carried out. In addition, one can also calculate the spin-projected Hessian (AP Hessian; F LS AP R ð Þ in Eq. (15)) as follows: where F LS BS and F HS are the Hessians calculated by the BS and the HS states, respectively. And a second derivative of α can be expressed by By using Eqs. (18) and (19), the spin-projected vibrational frequencies are also calculated. The AP optimization can be carried out based on Eq.
=∂R obtained by numerical fitting or analytical ways.

Relationship between the BS and projected wavefunctions
As well as a calculated energy and its derivatives, the BS wavefunction itself has also vital information. Here let us go back to Eq. (3). From the equation, an overlap between up-spin (socalled alpha) and down-spin (so-called beta) orbitals (T) becomes And because occupation number (n) of natural orbital (NO) for the corresponding orbital is expressed as n ¼ 2cos 2 θ, we get the relation: On the other hand, we can define projected wavefunction (PUHF) by eliminating triplet species from BS singlet wavefunction from Eq. (3) as follows: If we focus on the second term, which is related to double (two-electron) excitation, its weight (W D ) can be obtained from Eqs. (21) and (22) as follows: This is the weight of double excitation calculated by the BS wavefunction. By applying Eq. (21)-Eq. (23), the W D is related to the occupation number of the corresponding NO as follows: Approximate Spin Projection for Broken-Symmetry Method and Its Application http://dx.doi.org/10.5772/intechopen.75726 This y value is called an instability value of a chemical bond (or diradical character). In the case of the spin-restricted (or spin-adapted (SA)) calculations, the y value is zero. However if a couple of electrons tends to be separated and to be localized on each hydrogen atom, in other words the chemical bond becomes unstable with the strong static correlation effect, the y value becomes larger and finally becomes 1.0. So, the y value can be applied for the analyses of di-or polyradical species, and it is often useful to discuss the stability (or instability) of chemical bonds. The idea is also described by an effective bond order (b), which is defined by the difference in occupation numbers of occupied NO (n) and unoccupied NO (n * ): Different from the y value, the b value becomes smaller when the chemical bond becomes unstable. If we define the effective bond order with the spin projection b(AP), it is related to the y value: Those indices show how the BS and AP wavefunctions are connected. In addition, one can utilize the indices to estimate the contribution of double excitation for very large systems that CAS and MR methods cannot be applied.
However, in the case of the BS singlet state of H 2 molecule, it becomes Here we explain another aspect of the spin projection method. As depicted in Figure 1(c), the BS wavefunction indicates only one spin-polarized configuration, e.g., BS1 in the figure.
However, in order to obtain a pure singlet wavefunction, which satisfies the spin symmetry, the opposite spin-polarized state (BS2) must be included. The projection method can give a linear combination of the both BS states, and therefore it can give an appropriate quantum state for the singlet state.
3. Application of BS and AP methods to several biradical systems 3.1. Hydrogen molecule: comparison among SA, BS, and AP methods by simple biradical system In this section, we briefly illustrate how the BS and AP methods approximate a dissociation of a hydrogen molecule. Figure 2(a) shows potential energy curves of Hartree-Fock and full CI methods. In the case of the spin-adapted (SA) HF, i.e., the spin-restricted (R) HF method, the curve does not converge to the dissociation limit. On the other hand, the BS HF, i.e., spinunrestricted (U) HF calculation, successfully reproduces the dissociation limit of full CI method. This result indicates that the static correlation is included in the BS procedure. Around 1.2 Å, there is a bifurcation point between RHF and UHF methods. Within the closed shell (i.e., SA) region, where r H-H < 1.2 Å, the UHF solution does not appear, and the singlet state is described by RHF (single slater determinant). In this region, the energy gap between full CI and RHF that is known as correlation energy indicates a necessity of the dynamical correlation correction as discussed later.
In order to elucidate how the double-excitation state is included in the BS solution, the occupation numbers of the highest occupied natural orbital (HONO) are plotted along the H-H distance in Figure 2 Figure 3(a). This complex involves a quadruple Cr(II)-Cr(II) bond (σ, π // , π ⊥ , and δ orbitals). Due to the strong static correlation effect, it requires the multi-reference approach.
Within the BS procedure, as a consequence, the electronic structure of the complex is expressed by the spin localization on each Cr(II) ions. First, let us examine the nature of the metal-metal bond between Cr(II) ions. For the purpose, natural orbitals and their occupation numbers are obtained from the BS wavefunctions using an experimental geometry.
As depicted in Figure 3(b), there are eight magnetic orbitals, i.e., bonding and antibonding σ, π // , π ⊥ , and δ orbitals that concern about the direct bond between Cr(II) ions. The NO analysis clarifies the nature of the Cr-Cr bond. If d-orbitals of two Cr(II) ions have sufficient overlap to form the stable covalent bond, the occupation numbers of each occupied orbital will be almost 2.0 (i.e., T is close to 1.0). As summarized in Table 1, however, those bonds show much smaller values. The occupation numbers of all of occupied σ, π, and δ orbitals are close to 1.0, indicating that electronic structure of the complex 1 is described by a spin-polarized spin structure like the biradical singlet state.
By substituting the obtained energies and b S  Table 2. In comparison with the experimental value, HF method underestimates the effective exchange interaction, while B3LYP method overestimates it. This result is quite similar to a tendency of the J ab curve of H 2 molecule at the intermediation region in Figure 2(d). In that region, BH and HLYP method, which involves 50% HF exchange, gives better value in comparison with B3LYP. The results also suggest an importance of the effect on the ratio of the HF/DFT exchange for estimation of the effective exchange interaction [71,72].

Singlet methylene molecule: Spin contamination error in optimized geometry by BS method and its elimination by AP method
Finally, we examine the spin contamination error in the optimized structure. Here we focus on a singlet methylene (CH 2 ). As illustrated in Figure 4(a), the methylene molecule has two valence Approximate Spin Projection for Broken-Symmetry Method and Its Application http://dx.doi.org/10.5772/intechopen.75726 orbitals (ψ 1 and ψ 2 ) and two spins in those orbitals. Those two orbitals are orthogonal and energetically quasi-degenerate each other. The ground state of the molecule is 3 B 1 (triplet) state, and 1 A 1 (singlet) state is the first excited state. Components of the wavefunction of 1 A 1 state obtained by BS method as illustrated in Figure 4 (b) have been graphically explained [36]. The spin-restricted method such as RHF considers only single component (the first term of Figure 4(b)) although the BS wavefunction involves three components as illustrated in Figure 4 (b). The existence of the triplet component is the origin of the spin contamination error in this system. Both 1 A 1 and 3 B 1 methylene molecules have bent structures, but the experimental data indicates a large structural difference between them. For example, as summarized in Table 3, experimental HCH angles (θ HCH ) of 1 A 1 and 3 B 1 states are 102.4 and 134.0 , respectively [66,67]. There have also been many reports of the SA results as summarized in Ref. [68]. On the other hand, the BS method is a convenient substitute for CI and CAS method, so here we examined the optimized geometry of the 1 A 1 methylene by SA and BS methods. In order to elucidate a dependency of the spin contamination error on the calculation methods, HF, configuration interaction method with all double substitutions (CID), coupled-cluster method with double substitutions (CCD), several levels of Møller-Plesset energy correction methods (MP2, MP3, and MP4(SDQ)), and a hybrid DFT (B3LYP) method are also examined. In the case of 1   indicating that the post-HF methods even require some correction for such systems if the BS procedure is utilized. Therefore it is difficult to use the BS solution for 1 A 1 state without some corrections. On the other hand, by applying the AP method to the BS solution, the error is drastically improved, and the optimized structural parameters became in good agreement with experimental ones. The difference in the optimized θ HCH values between the BS and the AP method, i.e., the spin contamination error in the optimized geometry, is about 10-20 .   Those results strongly indicate that the spin contamination sometimes becomes a serious problem in the structural optimization of spin-polarized systems and the AP method can work well for its elimination. On the other hand, the optimized structure with the AP-UHF method almost corresponds to CASSCF (2,2) result. This means that the AP method approximates twoelectron excitation in the (2,2) active space well. The θ HCH values become smaller by including higher electron correlation with the larger CAS space such as CASSCF (6,6) or with the dynamical correlation correction such as MRMP2(2,2) and MRMP2 (6,6). The result of the spinprojected MP4 (AP MP4(SDQ)) successfully reproduced the MRMP2(6,6) result, indicating that the AP method plus dynamical correlation correction is a promising approach.
By calculating Hessian, one can also obtain frequencies of the normal modes. In Table 4, the calculated frequencies of the normal mode singlet methylene are summarized. The significant difference between the BS and AP methods can be found in a bending mode. The BS result underestimates the binding mode frequency by the contamination of the triplet state. On the other hand, the AP result gives close to the experimental result of 1 A 1 species. In this way, the AP method is also effective for the normal mode analysis as well as the geometry optimization.

Summary
In this chapter, we explain how the BS method breaks the spin symmetry and AP method recover it. In addition, we also demonstrate how those methods work the biradical systems. The theoretical studies of the large biradical and polyradical systems such as polynuclear metal complexes have been fairly realized by the BS HDFT methods in this decade. The BS method is quite powerful for the large degenerate systems, but one must be careful about the spin contamination error. Therefore the AP method would be important for those studies. For example, it is suggested that the spin contamination error misleads a reaction path that involves biradical transition states (TS) or intermediate state (IM) [73]. In addition, in the case of the more larger systems, e.g., metalloproteins, some kind of semiempirical approach combined with the AP hybrid DFT method by ONIOM method will be effective [74]. By using the method, the mechanisms of the long-distance electron transfers and so on will be elucidated. In  such cases, one also must be careful about the parameter of the semiempirical approach to fit the spin-polarized systems. Recently, some improvements for PM6 method have been proposed [75,76]. Because the PM6 calculation can be utilized for the outer region in ONIOM approach, therefore the AP method is also the effective method for the larger systems. In addition, the BS wavefunction can be applied for other molecular properties by combining with other theoretical procedures. For example, it was reported that the electron conductivity of spin-polarized systems could be simulated by using the BS wavefunction together with elastic Green's function method [77], and some applications for one-dimensional complexes have reported [78,79]. The results indicate that the BS wavefunctions can be applied for calculations of the physical properties of the strong electron correlation systems as well as their electronic structures. The spin-projected wavefunctions seem to be effective for such simulations of the physical properties. From those points of view, the BS and AP methods have a great potential to clarify chemical and physical phenomena that are still open questions.