betaFeedback

WorksPhysics and Astronomy

The tilted-plane structure of the energy of finite quantum systems

Abstract

The piecewise linearity condition on the total energy with respect to the total magnetization of finite quantum systems is derived, using the infinite-separation-limit technique. This generalizes the well-known constancy condition, related to static correlation error, in approximate density functional theory (DFT). The magnetic analog of the DFT Koopmans' theorem is also derived. Moving to fractional electron count, the tilted plane condition is derived, lifting certain assumptions in previous works. This generalization of the flat plane condition characterizes the total energy surface of a finite system for all values of electron count N and magnetization M. This result is used in combination with tabulated spectroscopic data to show the flat plane-structure of the oxygen atom, among others. We find that derivative discontinuities with respect to electron count sometimes occur at non-integer values. A diverse set of tilted plane structures is shown to occur in d-orbital subspaces, depending on chemical coordination. General occupancy-based total-energy expressions are demonstrated thereby to be necessarily dependent on the symmetry-imposed degeneracies.

Density Functional Theory (DFT), a workhorse of computational chemistry and condensed matter physics [1, 2, 3], owes its success to the development of relatively efficient, reliable and accurate approximations to the exchange-correlation (XC) functional [4, 5, 6, 7, 8, 9, 10, 11]. These approximations can be developed by enforcing exact conditions and appropriate norms on a functional of given mathematical form [12]. Two well-known such exact conditions are the piecewise linearity condition with respect to electron count NN [13] and the constancy condition with respect to spin-magnetization (henceforth referred to as magnetization for concision) MM [14, 15] that, as we will discuss, is a special case of a more general linearity condition. The former condition states that the total ground state energy of a system with external potential v(r)v(\mathbf{r}) and electron count N=N0+ωN=N_0+\omega is given by

Ev(N0+ω)=(1−ω)Ev(N0)+ωEv(N0+1),(1)E_v(N_0+\omega)=(1-\omega)E_v(N_0)+\omega E_v(N_0+1), \tag*{(1)}

where N0∈N0N_0\in\mathbb{N}^0 and 0<ω<10<\omega<1 [13, 14, 16]. A DFT calculation for a finite system with a non-integer electron count necessitates fractional occupancy of at least one Kohn Sham (KS) orbital. Assuming this fractional occupancy is limited to one KS orbital, the slope of the linear segment of the Ev(N)E_v(N) curve is given by

(∂Ev∂N)v=ϵf,(2)\left(\frac{\partial E_v}{\partial N}\right)_v=\epsilon_f, \tag*{(2)}

where ϵf\epsilon_f is the fractionally occupied KS eigenvalue [17, 18, 19, 20]. A derivative discontinuity must therefore occur at integer values of electron count [21, 22, 23, 24, 25, 26], at least. The left-hand partial such derivative is given by the highest occupied KS eigenvalue. The right-hand partial derivative is given by the lowest unoccupied KS eigenvalue, ϵLUKS\epsilon_{\mathrm{LUKS}}, with an additional contribution from the derivative discontinuity of the exact XC functional, that is by

lim⁡δ→0+(∂Ev∂N)v∣N0+δ=ϵLUKS+ΔxcN.(3)\lim_{\delta\to0^+}\left(\frac{\partial E_v}{\partial N}\right)_{v\mid N_0+\delta}=\epsilon_{\mathrm{LUKS}}+\Delta_{\mathrm{xc}}^N. \tag*{(3)}

The piecewise linearity condition with respect to electron count [13], as given by Eq. (1), follows from the convexity condition [27], namely that

2Ev[N0]≤Ev[N0−1]+Ev[N0+1].(4)2E_v[N_0]\le E_v[N_0-1]+E_v[N_0+1]. \tag*{(4)}

The second of the two aforementioned exact conditions is the constancy condition with respect to magnetization MM [14, 15], which states that the total ground state energy of a system with electron count N0N_0 and magnetization MM satisfies, for any magnetization ∣M∣≤M0\lvert M\rvert\le M_0,

Ev(N0,M)=Ev(N0,M0)=Ev(N0,−M0).(5)E_v(N_0,M)=E_v(N_0,M_0)=E_v(N_0,-M_0). \tag*{(5)}

Here, M0∈N0M_0\in\mathbb{N}^0 is the maximum magnetization of the lowest-energy state for a given integer electron count N0∈N0N_0\in\mathbb{N}^0. Ev(N,M)E_v(N,M) denotes the ground state energy of the system with specified electron count NN and magnetization MM, noting that these are independent parameters in DFT, subject to ∣M∣≤N\lvert M\rvert\le N. Here, and in what follows, it is supposed that no ambient field coupling to spin is present.

The linearity and constancy conditions given by equations 1 and 5 have been combined and generalized [14, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37] to give the flat plane condition

Ev(N,M)=(1−ω)Ev(N0,M0)+ωEv(N0+1,M1)\begin{aligned} E_v(N,M)={}&(1-\omega)E_v(N_0,M_0)\\ &+\omega E_v(N_0+1,M_1) \end{aligned}

for ∣M∣≤M0+ω(M1−M0)\lvert M\rvert\le M_0+\omega(M_1-M_0), where N=N0+ωN=N_0+\omega and M1∈N0M_1\in\mathbb{N}^0 is the maximum magnetization of the lowest-energy state for the N0+1N_0+1 electron system.

Many approximate XC functionals violate these exact conditions, and this failure has been directly linked to poor performance in the prediction of band gaps [38, 39, 40], molecular dissociation [41] [42, 43, 44], and electronic transport [45]. A small number of functionals enforce, fully or to some extent, the flat plane condition [37] [46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56].

It has long been recognised in the literature that the constancy condition with respect to magnetisation and the accompanying flat-plane condition apply only to the aforementioned limited interval of magnetization states. In previous works to generalize this, X. Yang et al. [36] proposed the existence of two types of flat planes, for example, and Gál and Geerlings [34] proposed a piecewise linearity condition for magnetization by invoking a zero-temperature grand canonical ensemble.

In this Letter, we rigorously classify the Ev[N,M]E_v[N,M] surface for all values of magnetization MM, as opposed to the limited interval ∣M∣≤M0+ω(M1−M0)\lvert M\rvert\le M_0+\omega(M_1-M_0), across all formulations of DFT that satisfy three minimum requirements. These are that the total energy functional is (a) exact for all vv-representable spin-densities, (b) size-consistent, which is connected to the idea of near-sightedness, and (c) translationally invariant. We derive a piecewise linearity condition with respect to magnetisation (Thm. 1) and an accompanying, general ‘tilted plane’ condition (Thm. 2). A tilted plane may have a non-zero partial derivative with respect to both NN and MM, while a flat plane (as defined by Eq. 6) can only have a non-zero partial derivative with respect to NN. In certain cases only, the tilted plane simplifies to the well-known flat plane.

Thms. 1 & 2 are exact conditions for the XC functional, and so may inform and stimulate the development of improved XC approximations. They highlight the opportunity for functionals that incorporate explicit derivative discontinuities with respect to electron count NN and magnetization MM. Thm. 1 opens a new avenue for incorporating spin-polarised atomic reference data as guidance for spin-polarized functional generalizations [57–59]. Furthermore, these two theorems demonstrate that occupancy-based functionals, such as in DFT+U type methods [46, 47, 49, 60–63] and certain machine-learning based methods [64–66], might beneficially take on different analytical expressions depending on symmetry-imposed degeneracies, as we approach exactitude.

In SM IV, Corollary 1.1 provides the analogue of the DFT Koopmans’ condition for spin, i.e., an exact condition on the spin-splitting of the effective (e.g., Kohn-Sham) potential. In spin-polarized systems, this splitting underpins the prediction of magnetic excitations and exchange couplings from a DFT starting point.

Thm. 2 classifies the Ev[N,M]E_v[N,M] curve for all values of magnetisation MM. Higher-energy magnetisation states are the subject of inherent interest and intense research. For example, the lowest energy triplet state plays a crucial role in phosphorescence [67, 68], thermally activated delayed fluorescence [69, 70] and singlet fission [71, 72]. Within spin-DFT, such higher energy magnetization states are accessible, as one may compute the lowest energy state of each symmetry [73].

Theorem 1. The Ev[N0,M]E_v[N_0,M] curve is piecewise linear with respect to magnetization MM.

The structure of the Ev(N0,M)E_v(N_0,M) curve (where N0∈NN_0 \in\mathbb{N}) can be elucidated by employing the technique, developed in W. Yang et al. [14], of constructing a system with external potential v(r)v(\mathbf{r}) that is composed of qq copies of the same finite system with external potential vR(r)v_{\mathrm{R}}(\mathbf{r}), with all copies being infinitely separated in space. The duplication of the system, at infinite separation, is a theoretical device used to consider its possible fractional charge and spin, without resorting to an ensemble treatment. The proof proceeds by analysing the overall system of duplicates, however it is the individual finite sub-system (which may be strongly multi-reference in character) with which we are ultimately concerned. We have that

v(r)=∑l=1qvRl(r).(6)v(\mathbf{r}) = \sum_{l=1}^{q} v_{\mathrm{R}l}(\mathbf{r}). \tag*{(6)}

The total electron count is denoted by qN0qN_0, where N0∈NN_0 \in\mathbb{N}. It follows from the convexity condition that, at the infinite separation limit, N0N_0 electrons will localize on each of the qq sites. For brevity, we will not repeat the electron count constraint ∣M∣≤N0∈N0\lvert M\rvert\le N_0 \in\mathbb{N}_0, but note that it holds throughout Thm. 1 and Cor. 1.1. The total magnetization of the system is denoted by qMqM, where qq and qM∈ZqM \in\mathbb{Z}. Typically, M∉ZM \notin\mathbb{Z} but M∈QM \in\mathbb{Q} (which is supposed to be sufficiently dense in R\mathbb{R} for present purposes). The ground state of the system will be composed of pp sites with a magnetization MiM_i, and q−pq-p sites with a magnetization MjM_j, where

qM=pMi+(q−p)Mj,p,q∈N0,qM = pM_i + (q-p)M_j,\qquad p,q \in\mathbb{N}^0,
p≤q,Mi,Mj∈Z,andMi≤M≤Mj.(7)p \le q,\qquad M_i,M_j \in\mathbb{Z},\qquad\text{and}\qquad M_i \le M \le M_j. \tag*{(7)}

For fixed values of the external parameters N0N_0 and MM, the total number of sites qq is chosen, together with the minimum MiM_i and maximum MjM_j, such that the total energy per site is minimized while satisfying the constraints of Eq. (7). Often one finds Mi=Mj−2M_i=M_j-2, but not always. For example, in the nitrogen atom Mi=−3M_i=-3 and Mj=+3M_j=+3 for −3<M<+3-3<M<+3. The choice of MiM_i and MjM_j is discussed further in SM I.

One of the degenerate ground state wave functions of the system will consist of pp sites with magnetisation MiM_i and q−pq-p sites with magnetisation MjM_j. As we describe in detail in SM II, we may take the average of all such ground state wave functions, and analyse the associated spin-resolved site electron density. The piecewise linearity condition for magnetisation, i.e.,

Ev(N0,M)=ωEv(N0,Mi)+(1−ω)Ev(N0,Mj),E_v(N_0,M)=\omega E_v(N_0,M_i)+(1-\omega)E_v(N_0,M_j),
M=ωMi+(1−ω)Mj,Mi,Mj∈Z & 0≤ω≤1.(8)M=\omega M_i+(1-\omega)M_j,\qquad M_i,M_j\in\mathbb{Z}\ \&\ 0\le\omega\le1. \tag*{(8)}

thereby follows for all total energy functionals that are (a) exact for all vv-representable spin-densities, (b) size-consistent, and (c) translationally invariant.

We note that Gál and Geerlings [34] arrived at a similar piecewise linearity condition for magnetization by invoking a zero-temperature grand canonical ensemble.

Approximate total energy functionals EaprxE^{\mathrm{aprx}} typically do not obey the piecewise linearity condition with respect to magnetization. We may refer to the intrinsic deviation of EaprxE^{\mathrm{aprx}} from the piecewise linear Ev(N0,M)E_v(N_0,M) curve, between the (often inexact) energies at integer magnetization, as magnetic piecewise linearity error (MPLE),

EMPLE(N0,M)=Evaprx(N0,M)−[ωEvaprx(N0,Mi)+(1−ω)Evaprx(N0,Mj)].(9)E^{\mathrm{MPLE}}(N_0,M)=E_v^{\mathrm{aprx}}(N_0,M)-[\omega E_v^{\mathrm{aprx}}(N_0,M_i)+(1-\omega)E_v^{\mathrm{aprx}}(N_0,M_j)]. \tag*{(9)}

Here the electron count is a constant integer value N0N_0 and MM, MiM_i, MjM_j and ω\omega are given by equation S11. Figure 1 shows the EMPLE(M)E^{\mathrm{MPLE}}(M) for the neutral helium atom.

Image not extracted

Figure 1. The magnetic piecewise linearity error (MPLE) for the neutral Helium atom exhibited by a variety of local [74], semi-local [5] [75], hybrid [7, 8, 9] [76] and van der Waals corrected [77] XC functionals, as well as Hartree-Fock (HF). As defined in Eq. (9), the MPLE for each approximate functional vanishes at certain integer values of spin-magnetization. MLPE exhibits a substantial cubic component for all functionals shown. It approximately vanishes near full magnetization (but not elsewhere) for the hybrid functionals, reflecting the negative sign of MPLE for HF in this test system and hence cancellation of error [78]. Calculation details [79–81] can be found in SM III.

MPLE differs from static correlation error (SCE) [15] [82], which is defined as the spurious difference in total energy, due to the use of an approximate XC functional, between states that should be degenerate. MPLE instead is an error in the total energy of non-integer magnetization states, irrespective of whether such a state should be degenerate with one with integer magnetization. It would not be so, typically, in tilted-plane regions, or in a finite external magnetic field. Conversely, not all SCE can be described as MPLE, because SCE may refer to spurious energy differences between states of the same magnetization. An example of an SCE that is not an MPLE is the energy error that may arise in the description of the spherically symmetric boron atom [15].

Corollary 1.1. The partial derivative of the Ev[N0,M]E_v[N_0,M] curve with respect to magnetization MM is equal to half the difference between the spin-dependent frontier KS eigenvalues ϵfσ\epsilon_f^\sigma, whenever the partial derivative exists,

(∂Ev∂M)N=12[ϵf↑−ϵf↓].(10)\left(\frac{\partial E_v}{\partial M}\right)_N=\frac{1}{2}\left[\epsilon_f^\uparrow-\epsilon_f^\downarrow\right]. \tag*{(10)}

A simple proof of this corollary is given in SM IV (see also references [24, 28, 34] [83, 84, 85, 86] therein), which avoids the need to invoke total single-particle energies or grand canonical ensembles.

Theorem 2. The Ev[N,M]E_v[N,M] surface obeys the tilted plane condition described by Eq. (13).

Analysis of the Ev[N0,M]E_v[N_0,M] curve may be extended to include states with not only fractional magnetization MM but also fractional electron count NN. Again, one may construct an external potential given by Eq. (6), in this case typically both NN and M∉ZM \notin\mathbb{Z}, such that the total system electron count and magnetization, qNqN and qM∈ZqM \in\mathbb{Z}. The ground state of this system will be composed of qiq_i sites with electron count NiN_i and magnetization MiM_i, where ii ranges from 1 to VcountV_{\mathrm{count}} with:

∑i=1Vcountci=1,0≤ci≤1,Ni, Mi, qci∈Z.(11)\sum_{i=1}^{V_{\mathrm{count}}} c_i=1,\qquad0\le c_i\le1,\qquad N_i,\ M_i,\ qc_i\in\mathbb{Z}. \tag*{(11)}

The values of cic_i, NiN_i and MiM_i are constrained so that

N=∑i=1VcountciNiandM=∑i=1VcountciMi.(12)N=\sum_{i=1}^{V_{\mathrm{count}}}c_iN_i\quad\text{and}\quad M=\sum_{i=1}^{V_{\mathrm{count}}}c_iM_i. \tag*{(12)}

Following an analogous derivation to that outlined in Thm. 1 (inc. SM I and II), one finds that, for N,M∉ZN,M\notin\mathbb{Z},

Ev[N,M]=∑i=1VcountciEv[Ni,Mi].(13)E_v[N,M]=\sum_{i=1}^{V_{\mathrm{count}}}c_iE_v[N_i,M_i]. \tag*{(13)}

Vertices in the energy landscape will occur at the specified integer values of electron count NiN_i and magnetization MiM_i. VcountV_{\mathrm{count}} is the number of vertices associated with a particular plane, typically equal to 3 or 4, however higher numbers of vertices are possible in very rare circumstances, as discussed in SM V. To simplify discussion, we restrict Vcount=4V_{\mathrm{count}}=4, however, the method used to generate Figure 2 includes no restriction on VcountV_{\mathrm{count}}.

Figure 2. The projection of the Ev[N,M]E_{\mathrm{v}}[N,M] surface of the oxygen, chromium, manganese and iron atoms onto the N−MN-M plane. The Ev[N,M]E_{\mathrm{v}}[N,M] curves are composed of a series of three- and four-sided planes with vertices at integer values of NN and MM. (The vertical axis is labelled in terms of the total atomic charge as opposed to total electron count.) The total energy varies linearly across each plane. Each plane is outlined in black. Energy contour lines are plotted at intervals of 100.1x10^{0.1\mathrm{x}} eV. For each atom, the total energy values are based on available experimental NIST reference data [87] and are given relative to the ground state total energy of the neutral atom in its lowest energy magnetization state, which is set to 1 eV for visualization purposes. Orbital and nuclear moments are not included. The Ev[N,M]E_{\mathrm{v}}[N,M] surface of the oxygen atom for charged states between +5+5 and +6+6 is composed of a series of five triangular shaped planes, four of which can neither be classed as a type 1 or a type 2 flat plane of X. Yang et al. [36]. Note that planes at the edges are omitted due to absence of experimental reference data. If any additional experimental reference data points not currently included are found to be low enough in energy, they could also affect the shape of the outlying, high-magnetization Ev[N,M]E_{\mathrm{v}}[N,M] surfaces shown here.

In cases where the electron counts of the four states at the vertices satisfy N1=N2=N3−1=N4−1N_1=N_2=N_3-1=N_4-1 and the magnetizations M1=−M2M_1=-M_2, M3=−M4M_3=-M_4, with Ev[Ni,Mi]=min⁡M{Ev[Ni,M]}E_v[N_i,M_i]=\min_M\{E_v[N_i,M]\}, Eq. (13) simplifies to the flat-plane condition as outlined in Eq. 6. X. Yang et al. [36] and Cuevas-Saavedra et al. [37] report the existence of two types of flat plane structures. These types are special cases (non-exhaustive ones; see Figure 2 for an exception) of the more general condition outlined by Eq. (13), specifically when c4=0c_4=0. Generally, restricting c4c_4 to be zero prohibits the correct flat plane structure of systems, e.g., the oxygen atom for 7≤N≤87 \leq N \leq8 and ∣M∣≤10−N\lvert M\rvert\leq10-N, as highlighted in white in Figure 2. The correct expression for this energy surface is detailed in SM VI, along with further analysis of Figure 2.

Gál and Geerlings [34] reported the existence of a tilted plane energy surface but their energy expressions also have the c4=0c_4=0 restriction, meaning that (Ni,Mi)(N_i,M_i) values in their energy expression will not always represent vertices in the energy landscape. The same is true for Chan’s Ev[N,M]E_{\mathrm{v}}[N,M] energy expression [28]. To the best of our knowledge, a lifting of the c4=0c_4=0 restriction of the generalized flat plane condition has only been discussed to date in the unpublished Ref. [35]. If we assume that (Ni,Mi)(N_i,M_i) values represent vertices in the energy landscape, the c4=0c_4=0 restriction allows for triangular shaped planes but neglects planes of other shapes, such as isosceles trapezoids, which occur in the oxygen atom (Figure 2).

A consequence worthy of future investigation is the appearance of derivative discontinuities both at non-integer values of total electron count NN, and of spin-electron count 12(N+M)\frac{1}{2}(N+M), as seen, e.g., in the oxygen atom between charge states 5+5+ and 6+6+ with 1≤M≤21 \leq M \leq2 (Figure 2). Knowledge of the occurrence of a tilted plane energy surface has already been applied to correct the PBE total energy of dissociated triplet H2+\mathrm{H}_2^+ in Ref. [46]. An analysis of the tilted plane shape of the Ev[N,M]E_{\mathrm{v}}[N,M] surface for the He atom is shown in SM III and the Ev[N,M]E_{\mathrm{v}}[N,M] surface for the multi-reference, quintuple bonded VCr, VMo and CrMo hetero-dimers [88] are presented in SM VII. Sample degeneracy-dependent tilted plane structures for an isolated dd-orbital subspace are shown in SM VIII.

In conclusion, the piecewise linearity condition with respect to magnetization and the tilted plane condition were derived from first principles using the infinite-separation-limit technique, and the magnetic analogue of the DFT Koopmans’ theorem was derived from the chain rule. These exact conditions have been derived for all formulations of DFT that are exact for all vv-representable spin-densities, size-consistent, and translationally invariant. It is important to note that these exact conditions apply only to the lowest Ev[N,M]E_v[N, M] energy surface. By performing a constrained search over states of a specified spin multiplicity [73], it is possible to access other quantum states within DFT that are not located on this energy surface, for example, the lowest energy singlet state of a system with a triplet ground state.

We have found that many standard density functional approximations violate these exact conditions. These three exact quantum mechanical conditions may aid in the development of post-DFT methods and functional approximations, machine learning and alchemical approaches in both condensed matter physics and quantum chemistry, and error-correction techniques involving total energies in quantum science. We find that in order to approach the exact limit, energy functionals of occupancies must necessarily take different forms depending on symmetry-imposed degeneracies.

Acknowledgments

The research conducted in this publication was funded by the Irish Research Council under grant number GOIPG/2020/1454. EL gratefully acknowledges financial support from the Swiss National Science Foundation (grant number 213082). This research was supported by the NCCR MARVEL, a National Centre of Competence in Research, funded by the Swiss National Science Foundation (grant number 205602).

References

SM-I. The correct choice of site magnetizations

In the derivation of the piecewise linearity condition for magnetization, care must be taken in determining the site magnetizations MiM_i and MjM_j. Here we have given values of the external parameters N0N_0 and MM, where qN0qN_0 and qMqM are, respectively, the total electron count and the total magnetization of the system composed of qq infinitely separated sites. The minimum MiM_i and maximum MjM_j values which minimize the total energy per site, while satisfying the constraint of Eq. (7), are determined by the following three conditions.

  1. The site magnetizations MiM_i and MjM_j must satisfy

Mi≤M≤MjMi,Mj∈Z.(14)M_i \leq M \leq M_j \qquad M_i,M_j \in\mathbb{Z}. \tag*{(14)}
  1. For any values of site magnetization Mh,Mk∈ZM_h,M_k \in\mathbb{Z} where Mi∉{Mh,Mk}M_i \notin\{M_h,M_k\}, the ground-state energy of the system with magnetization MiM_i must satisfy

EvRL(Mi)<chEvRL(Mh)+ckEvRL(Mk),(15)E_{v_{\mathrm{R}_\mathrm{L}}}(M_i) < c_h E_{v_{\mathrm{R}_\mathrm{L}}}(M_h) + c_k E_{v_{\mathrm{R}_\mathrm{L}}}(M_k), \tag*{(15)}
Mi=chMh+ckMk,ch,ck≥0&ch+ck=1.M_i = c_h M_h + c_k M_k,\qquad c_h,c_k \geq0\quad\&\quad c_h+c_k=1.

The equivalent condition for MjM_j must also be true.

  1. For any integer value of site magnetization M′M' in the range Mi≤M′≤MjM_i \leq M' \leq M_j, the ground-state energy must satisfy

EvRL(M′)≥ciEvRL(Mi)+cjEvRL(Mj),(16)E_{v_{\mathrm{R}_\mathrm{L}}}(M') \geq c_i E_{v_{\mathrm{R}_\mathrm{L}}}(M_i) + c_j E_{v_{\mathrm{R}_\mathrm{L}}}(M_j), \tag*{(16)}
M′=ciMi+cjMj,ci,cj≥0&ci+cj=1.M' = c_i M_i + c_j M_j,\qquad c_i,c_j \geq0\quad\&\quad c_i+c_j=1.

Thus, the same values of MiM_i and MjM_j will satisfy these criteria for all values of the external parameter MM that satisfy Eq. (14).

SM-II. The Piecewise Linearity Condition for Magnetization

Given a system of qq infinitely separated sites with total electron count qN0qN_0 and magnetization qMqM, where the ground state of the system is composed of pp sites with a magnetization MiM_i, and q−pq-p sites with a magnetization MjM_j, subject to the constraints of Eq. (7) in the main text, we explicate in full the proof of Eq. (8) in the main text, which specifies the piecewise linearity condition for magnetization. The correct choice in values of MiM_i and MjM_j have already been discussed in SM-I.

The system has a set of degenerate ground state wave functions. At least one of these degenerate wave functions will be the antisymmetrized product of ground state wave functions at each site [14, 16]. This follows because in the infinite separation limit, each site cannot interact with its neighbours and, thus, minimizing the energy of the total system is equivalent to separately minimizing the energy associated with each site subject to the constraints of Eq. (7). One such ground state will have the first pp sites with magnetization MiM_i and wave function ΦMi\Phi_{M_i}, and the remaining q−pq-p sites with magnetization MjM_j and wave function ΦMj\Phi_{M_j}. The ground state wave function of the total system is

Ψ1=A^(ΦMi(R1)⋯ΦMi(Rp)\Psi_1 = \hat{A}\left(\Phi_{M_i}(\mathbf{R}_1)\cdots\Phi_{M_i}(\mathbf{R}_p)\right.
ΦMj(Rp+1)⋯ΦMj(Rq)).(17)\left.\Phi_{M_j}(\mathbf{R}_{p+1})\cdots\Phi_{M_j}(\mathbf{R}_q)\right). \tag*{(17)}

It is important to note that the ground state wave function of Eq. (17) holds regardless of whether each finite sub-system can be well-described by a single-determinant wave function or not.

Exchanging the magnetizations MiM_i and MjM_j at any two sites results in a degenerate ground state wave function. Therefore, the average of all such wave functions Ψavg\Psi_{\mathrm{avg}} is also a ground state wave function of the system. The ground state spin σ\sigma density of Ψavg\Psi_{\mathrm{avg}} is given by

ρσ(r)=∑l=1qpqρlσ(r;Mi)+q−pqρlσ(r;Mj),(18)\rho^\sigma(\mathbf{r}) = \sum_{l=1}^{q}\frac{p}{q}\rho_l^\sigma(\mathbf{r};M_i)+\frac{q-p}{q}\rho_l^\sigma(\mathbf{r};M_j), \tag*{(18)}

where ρlσ(r;Mi)\rho_l^\sigma(\mathbf{r};M_i) is the spin σ\sigma spin density of site ll with a magnetization of MiM_i. In this case, each of the pp (and separately q−pq-p) sites have identical spin-resolved densities. We emphasize that Eq. (18) is one of the degenerate ground state spin σ\sigma densities of the system with total magnetization qMqM, where qq is the number of sites and MM is limited to the interval Mi≤M≤MjM_i \leq M \leq M_j.

To deduce the piecewise linearity condition for magnetization, we make three reasonable assertions about the nature of the total energy functional, namely that it is (a) exact for all vv-representable spin-densities, (b) size-consistent, and (c) translationally invariant. We expect that the exact functional satisfies these in all DFT formulations where the ground state is generally accessible.

From (a), the total energy function should be exact for the spin resolved density of Eq. (18), so that

Ev[ρσ]=pEvRL(Mi)+(q−p)EvRL(Mj).(19)E_v[\rho^\sigma] = pE_{v_{\mathrm{R}_\mathrm{L}}}(M_i)+(q-p)E_{v_{\mathrm{R}_\mathrm{L}}}(M_j). \tag*{(19)}

From (b), the total energy functional should be size-consistent, whereupon

Ev(ρσ)=∑l=1qEvRL(pqρlσ(Mi)+q−pqρlσ(Mj)).(20)E_v(\rho^\sigma)=\sum_{l=1}^{q}E_{v_{\mathrm{R}_\mathrm{L}}}\left(\frac{p}{q}\rho_l^\sigma(M_i)+\frac{q-p}{q}\rho_l^\sigma(M_j)\right). \tag*{(20)}

Eq. (20) can be simplified by application of (c), translational invariance, following which

Ev[ρσ]=qEvRL[pqρσ(Mi)+q−pqρσ(Mj)].(21)E_v[\rho^\sigma]=qE_{v_{\mathrm{R}_\mathrm{L}}}\left[\frac{p}{q}\rho^\sigma(M_i)+\frac{q-p}{q}\rho^\sigma(M_j)\right]. \tag*{(21)}

From Eqs. (19) and (21), the total energy of the isolated site ll with magnetization MlM_l is given by

EvRl(N0,Ml)=pqEvRl(N0,Mi)+q−pqEvRl(N0,Mj),(22)E_{v_{\mathbf{R}_l}}(N_0,M_l)=\frac{p}{q}E_{v_{\mathbf{R}_l}}(N_0,M_i)+\frac{q-p}{q}E_{v_{\mathbf{R}_l}}(N_0,M_j), \tag*{(22)}

where the site magnetization is given by

Ml=pqMi+q−pqMj=M.(23)M_l=\frac{p}{q}M_i+\frac{q-p}{q}M_j=M. \tag*{(23)}

An equivalent argument will hold for any value of MM in the range Mi≤M≤MjM_i\leq M\leq M_j. Making the change of variable ω=p/q\omega=p/q and relabelling the site potential vRl(r)v_{\mathbf{R}_l}(\mathbf{r}) simply as v(r)v(\mathbf{r}), we may succinctly state that

Ev(N0,M)=ωEv(N0,Mi)E_v(N_0,M)=\omega E_v(N_0,M_i)
+(1−ω)Ev(N0,Mj),(24)\quad+(1-\omega)E_v(N_0,M_j), \tag*{(24)}
M=ωMi+(1−ω)Mj,Mi,Mj∈Z & 0≤ω≤1.M=\omega M_i+(1-\omega)M_j,\quad M_i,M_j\in\mathbb{Z}\ \&\ 0\leq\omega\leq1.

Therefore, a spin density functional Ev[ρ↑(r),ρ↓(r)]E_v[\rho^\uparrow(\mathbf{r}),\rho^\downarrow(\mathbf{r})] satisfying the three aforementioned conditions obeys a piecewise linearity condition with respect to magnetization, as opposed to simply a constancy condition.

Helium atom at Fractional Magnetization

It is instructive to apply our analysis on the helium energy surface at fractional values of NN and MM. The Ev[N,M]E_v[N,M] surface for the helium atom with electron count in the range 1≤N≤21\leq N\leq2 and magnetization M≥2−NM\geq2-N is given by

Ev[N,M]=(2−N)Ev[1,1]+N−M2Ev[2,0]+M+N−22Ev[2,2].(25)E_v[N,M]=(2-N)E_v[1,1]+\frac{N-M}{2}E_v[2,0] +\frac{M+N-2}{2}E_v[2,2]. \tag*{(25)}

This plane will have a non-zero partial derivative with respect to magnetization and thus has a ‘tilted plane’ shape as opposed to a simple ‘flat plane’ shape. Approximate XC functionals typically do not obey the tilted plane condition. For example, Figure S1 displays the energy of the helium atom when using the PBE XC functional, with respect to the exact tilted plane outlined in Eq. (25).

Deviation of the total energy using the PBE XC functional from the tilted plane condition

Figure S1. A plot of the deviation of the total energy using the PBE XC functional from the tilted plane condition for the Helium atomic system with spin up and spin down electron counts in the range 1≤n↑≤21\leq n^\uparrow\leq2 and 0≤n↓≤10\leq n^\downarrow\leq1, assuming the PBE total energy is exact at integer values of n↑n^\uparrow and n↓n^\downarrow.

All DFT calculations on the helium atom, as presented also in Figure 1 of the main text, were performed using the pw.x code of the Quantum Espresso package [79]. Kintetic-energy cutoffs of 75 Ry and 300 Ry were set for the Kohn Sham wavefunctions and charge density, respectively. Martyna-Tuckerman periodic boundary corrections [80] were applied, along with a Gaussian occupancy smearing of 0.02 Ry. Davidson diagonalization was employed with Broyden charge density mixing, a mixing factor of 0.7, and a convergence threshold of 5×10−95\times10^{-9} Ry. The helium atom was positioned in the centre of a

20×20×2020\times20\times20 Å3^3 cell, with Γ-point only reciprocal space sampling. The Optimized Norm-Conserving Vanderbilt helium pseudopotential [81] from the SG15 collection was used, which is available at www.quantum-simulation.org. In the case of Hartree-Fock, B3LYP, HSE and PBE0, the fully magnetized (M=2M=2) helium calculations would not converge. In this case, the total energy at M=2M=2 was evaluated by linear extrapolation of the total energies from the M=1.97M=1.97, 1.981.98 & 1.991.99 calculations.

The magnetic analogue of the DFT Koopmans’ theorem

Corollary 1.1. The partial derivative of the Ev[N0,M]E_{\mathrm{v}}[N_{0}, M] curve with respect to magnetization MM is equal to half the difference between the spin-dependent frontier KS eigenvalues ϵfσ\epsilon_{\mathrm{f}}^{\sigma}, whenever the partial derivative exists.

The magnetic analogue of the DFT Koopmans’ theorem can be derived by simple application of the chain rule in conjunction with the well studied (spin resolved) DFT Koopmans’ theorem. This dispenses with the need to invoke total single particle energies or grand canonical ensembles used in previous proofs [34] [84] of this theorem. The partial derivative of the total energy with respect to magnetization may be expressed in terms of the spin resolved electron counts

(∂Ev∂M)N=∑σ(∂Ev∂Nσ)Nσˉ(∂Nσ∂M)N,(26)\left(\frac{\partial E_{\mathrm{v}}}{\partial M}\right)_{N} = \sum_{\sigma}\left(\frac{\partial E_{\mathrm{v}}}{\partial N^{\sigma}}\right)_{N^{\bar{\sigma}}}\left(\frac{\partial N^{\sigma}}{\partial M}\right)_{N}, \tag*{(26)}

whenever the necessary partial derivatives exist. The right and left partial derivatives of EvE_{\mathrm{v}} with respect to NσN^{\sigma} are given, respectively, by

lim⁡δ→0+(∂Ev∂Nσ)N0σ+δ=ϵLUKSσ+ΔxcNσ\lim_{\delta\to0^{+}}\left(\frac{\partial E_{\mathrm{v}}}{\partial N^{\sigma}}\right)_{N_{0}^{\sigma}+\delta}=\epsilon_{\mathrm{LUKS}}^{\sigma}+\Delta_{\mathrm{xc}}^{N^{\sigma}}

and

lim⁡δ→0−(∂Ev∂Nσ)N0σ+δ=ϵHOKSσ,(27)\lim_{\delta\to0^{-}}\left(\frac{\partial E_{\mathrm{v}}}{\partial N^{\sigma}}\right)_{N_{0}^{\sigma}+\delta}=\epsilon_{\mathrm{HOKS}}^{\sigma}, \tag*{(27)}

where ϵHOKSσ\epsilon_{\mathrm{HOKS}}^{\sigma} is the highest occupied spin-σ\sigma KS eigenvalue and ϵLUKSσ\epsilon_{\mathrm{LUKS}}^{\sigma} is the lowest unoccupied spin-σ\sigma KS eigenvalue. ΔxcNσ\Delta_{\mathrm{xc}}^{N^{\sigma}} is the explicit derivative discontinuity of the exact XC functional through its explicit dependence on NσN^{\sigma}. Eq. (27) is the spin resolved analogue of Eqs. [(2)] and [(3)]. For further details, see Refs. [24, 28] [85, 86, 87].

The right and left partial derivatives of the total energy with respect to magnetization of a system with no fractional KS orbital occupations is thus given by

lim⁡M→M0+(∂Ev∂M)N=12[ϵLUKS↑−ϵHOKS↑+ΔxcN↑]\lim_{M\to M_{0}^{+}}\left(\frac{\partial E_{\mathrm{v}}}{\partial M}\right)_{N}=\frac{1}{2}\left[\epsilon_{\mathrm{LUKS}}^{\uparrow}-\epsilon_{\mathrm{HOKS}}^{\uparrow}+\Delta_{\mathrm{xc}}^{N^{\uparrow}}\right]

and

lim⁡M→M0−(∂Ev∂M)N=12[ϵHOKS↓−ϵLUKS↓−ΔxcN↓],(28)\lim_{M\to M_{0}^{-}}\left(\frac{\partial E_{\mathrm{v}}}{\partial M}\right)_{N}=\frac{1}{2}\left[\epsilon_{\mathrm{HOKS}}^{\downarrow}-\epsilon_{\mathrm{LUKS}}^{\downarrow}-\Delta_{\mathrm{xc}}^{N^{\downarrow}}\right], \tag*{(28)}

since the partial derivative of NσN^{\sigma} with respect to MM is equal to one half. For systems with one fractionally occupied spin up frontier KS orbital of eigenvalue ϵf↑\epsilon_{\mathrm{f}}^{\uparrow}, and one fractionally occupied spin down one of eigenvalue ϵf↓\epsilon_{\mathrm{f}}^{\downarrow}, the expression given by Eq. (28) simplifies to

(∂Ev∂M)N=12[ϵf↑−ϵf↓].(29)\left(\frac{\partial E_{\mathrm{v}}}{\partial M}\right)_{N}=\frac{1}{2}\left[\epsilon_{\mathrm{f}}^{\uparrow}-\epsilon_{\mathrm{f}}^{\downarrow}\right]. \tag*{(29)}

Typical XC functionals will break this exact condition. In Figure S2 the slope of the Ev[N0,M]E_{\mathrm{v}}[N_{0}, M] curve for the Helium atom, as evaluated from the PBE KS eigenvalues, is plotted as a function of magnetization. Use of the exact XC functional would result in a perfect step function.

Image not extracted

Figure S2. Half the difference of the fractionally occupied KS eigenvalues for the He atom, plotted as a function of magnetization MM, for a variety of local [74], semi-local [5] [75], hybrid [7, 8, 9] [76] and van der Waals corrected XC functionals [77], as well as Hartree-Fock (HF). The exact constant line, based on experimental NIST reference data [88], is also displayed. The lack of a derivative discontinuity within certain XC approximations can result in very large deviations from the magnetic analogue of Koopmans’ theorem at certain integer values of MM, as is the case shown here for the PBEsol, PBE and RVV10 functionals at M=2.0M = 2.0. As discussed in SM-III, the hybrid functionals as well as Hartree-Fock were evaluated at M=1.97M = 1.97, 1.981.98 & 1.991.99 but not at M=2.0M = 2.0.

Cases of high vertex count

In special cases, the vertex count of a given plane can be higher than four. For example, let us assume that the maximum magnetization of the lowest energy state of the N−1N - 1, NN and N+1N + 1 electron systems are given by MM, M+1M + 1, and MM, respectively, where M≥1M \ge1. It is assumed that the convexity condition of Eq. (4) is satisfied for the N−1N - 1, NN and N+1N + 1 electron systems. In the special case where

Ev[N,M+1]−Ev[N−1,M]N−(N−1)=Ev[N+1,M]−Ev[N,M+1](N+1)−N,(30)\begin{aligned} \frac{E_{\mathrm{v}}[N,M+1]-E_{\mathrm{v}}[N-1,M]}{N-(N-1)} \\ &=\frac{E_{\mathrm{v}}[N+1,M]-E_{\mathrm{v}}[N,M+1]}{(N+1)-N}, \tag*{(30)} \end{aligned}

the plane in question will have an hexagonal shape (stretched in the magnetization direction whenever $M > Such a plane would require the summation in Eq. (13) to be extended over six vertices

Ev[N,M]=∑i=16ciEv[Ni,Mi].(31)E_v[N,M] = \sum_{i=1}^{6} c_i E_v[N_i,M_i]. \tag*{(31)}

Higher numbers of vertices will occur when there is a fortuitous equality of certain energy derivatives. Eq. (30) specifies the energy derivative equality for this particular example. It is worth noting that Eq. (30) simplifies to

2Ev[N,M+1]=Ev[N−1,M]+Ev[N+1,M],(32)2E_v[N,M+1] = E_v[N-1,M] + E_v[N+1,M], \tag*{(32)}

however this simplification will not occur in all cases of high vertex count.

SM-VI. Further Analysis of Figure 2

A number of other interesting observations can be made from Figure 2 in the main text, for simplicity we have included the same figure here, labelled Figure S3. If we look, for example, at the 5+ oxygen anion, we can read from the graph that the energy is minimized at M=1M = 1 (the spin-doublet configuration 1s22s11s^2 2s^1), with a piecewise constant segment for ∣M∣≤1|M| \le1, while the next lowest magnetization state is M=3M = 3 (the spin-quartet configuration 1s12s1(3S)2p11s^1 2s^1({}^{3}S)2p^1), with piecewise linear (not piecewise constant) segments connecting these for 1≤∣M∣≤31 \le|M| \le3.

Figure S3. The projection of the Ev[N,M]E_{\mathrm{v}}[N,M] surface of the oxygen, chromium, manganese and iron atoms onto the N−MN-M plane. The Ev[N,M]E_{\mathrm{v}}[N,M] curves are composed of a series of three- and four-sided planes with vertices at integer values of NN and MM. (The vertical axis is labelled in terms of the total atomic charge as opposed to total electron count.) The total energy varies linearly across each plane. Each plane is outlined in black. Energy contour lines are plotted at intervals of 100.1x eV10^{0.1x}\,\mathrm{eV}. For each atom, the total energy values are based on available experimental National Institute of Standards and Technology reference data [88] and are given relative to the ground state total energy of the neutral atom in its lowest energy magnetization state, which is set to 1 eV for visualization purposes. Orbital and nuclear moments are not included. The Ev[N,M]E_{\mathrm{v}}[N,M] surface of the oxygen atom for charged states between +5 and +6 is composed of a series of five triangular shaped planes, four of which can neither be classed as a type 1 or a type 2 flat plane of X. Yang et al. [36]. Within this region, the energy surface has been incorrectly produced by Gál and Geerlings in Fig. 3 of their work [34], highlighting the challenge of mapping these planes. Note that planes at the edges are omitted due to absence of experimental reference data. If any additional experimental reference data points not currently included are found to be low enough in energy, they could also affect the shape of the outlying, high-magnetization Ev[N,M]E_{\mathrm{v}}[N,M] surfaces shown here. The flat plane described by Eq. (33) is highlighted in white.

It is also worth noting that the neutral chromium atom violates the Madelung rule, with a ground state electron configuration of [Ar]3d54s1[\mathrm{Ar}]3d^5 4s^1. This can been seen in the central uppermost plane of the Ev[N,M]E_v[N,M] surface for Cr, which extends from −6-6 to +6+6 at zero charge as opposed to −4-4 to +4+4, which would be the case if the configuration followed the Madelung rule ([Ar]3d44s2[\mathrm{Ar}]3d^4 4s^2).

The correct energy expression for the oxygen atom, without the c4=0c_4 = 0 restriction, may be written as

Ev[N,M]=c1Ev[8,2]+c2Ev[8,−2]+c3Ev[7,3]+c4Ev[7,−3]=(N−7)(10−N+M)2(10−N)Ev[8,2]+(N−7)(10−N−M)2(10−N)Ev[8,−2]+(8−N)(10−N+M)2(10−N)Ev[7,3]+(8−N)(10−N−M)2(10−N)Ev[7,−3],(33)\begin{aligned} E_v[N,M] &= c_1E_v[8,2] + c_2E_v[8,-2] + c_3E_v[7,3] + c_4E_v[7,-3] \\ &= \frac{(N-7)(10-N+M)}{2(10-N)}E_v[8,2] + \frac{(N-7)(10-N-M)}{2(10-N)}E_v[8,-2] \\ &\quad+ \frac{(8-N)(10-N+M)}{2(10-N)}E_v[7,3] + \frac{(8-N)(10-N-M)}{2(10-N)}E_v[7,-3], \tag*{(33)} \end{aligned}

for all 7≤N≤87 \le N \le8 and ∣M∣≤10−N|M| \le10-N. The flat plane described by Eq. (33) is highlighted in white in Figure S3.

Despite Ev[8,2]=Ev[8,−2]E_v[8,2] = E_v[8,-2] and Ev[7,3]=Ev[7,−3]E_v[7,3] = E_v[7,-3] for the oxygen atom, reduction of Eq. (33) to the sum of two terms with coefficients c1′c'_1 and c2′c'_2 would wrongly give

M≠∑i=12ci′Mi.(34)M \ne\sum_{i=1}^{2} c'_i M_i. \tag*{(34)}

The black lines within the Ev[N,M]E_v[N,M] surfaces of Figure 2 (Figure S3), correspond to values of NN & MM where derivative discontinuities in the total energy occur. As shown in the Ev[N,M]E_v[N,M] surface of the oxygen atom, these derivative discontinuities do not solely occur at integer values of NN & MM, or at integer values of the spin up and spin down electron counts (N↑N^\uparrow and N↓N^\downarrow).

For example, there is a line segment stretching from the point (0,6)(0,6) to the point (3,5)(3,5). This line represents a ‘seam’ which, in this case, both of the partial derivatives change. Thus, there is a derivative discontinuity in the total energy for all values of total charge C=8−NC = 8-N (8 being the atomic number of oxygen) and total magnetization MM on this line segment. The equation of the line segment is C=−13M+6C = -\frac{1}{3}M + 6, and so derivative discontinuities occur, e.g., at M=1.5M = 1.5 and C=5.5C = 5.5 as well as at the integer magnetization M=2M = 2 and C=513C = 5\frac{1}{3}. This discontinuity (for varying NN at constant MM) has a magnitude of approximately 10.2 eV.

This example also provides that a derivative discontinuity may be encountered while one spin occupancy alone is varied, at possibly non-integer values. Recall that the spin-indexed occupancies are given by N↑=(N+M)/2N^\uparrow= (N+M)/2 and N↓=(N−M)/2N^\downarrow= (N-M)/2. We may consider traversing from point (2,6)(2,6) (which corresponds to N↑=2N^\uparrow= 2 and N↓=0N^\downarrow= 0) to point (1,5)(1,5) (which corresponds to N↑=2N^\uparrow= 2 and N↓=1)N^\downarrow= 1). A derivative discontinuity is encountered at half-integer filling N↓=0.5N^\downarrow= 0.5, which is at the aforementioned point (1.5,5.5)(1.5,5.5), half-way along the line segment. It is worth noting that there appears to be no possible absent NIST spectroscopic data in the region in question that might correspond to higher-energy spin states, the existance of which might otherwise change the fracturing of the Ev[N,M]E_v[N,M] landscape and affect this prediction.

SM-VII. Ev[N,M]E_v[N,M] surface of sample transition metal hetero-dimers

The projection of the Ev[N,M]E_v[N,M] surface onto the N−MN-M plane for the strongly correlated CrMo, VCr and VMo transition metal hetero-dimers are presented in Figure S4. In particular, please note that the Ev[N,M]E_v[N,M] surfaces are composed of a series of three and four sided planes with vertices at integer values of NN and MM, as is the case with the oxygen, chromium, manganese and iron atoms. The total energy varies linearly across each plane but, for ease of visualization, each plane is colored based on the average of the energies at the vertices. The total energy values for the strongly correlated CrMo, VCr and VMo transition metal hetero-dimers are based on CASSCF/CASPT2 reference calculations on vertical excitations from Ref. [89]. This study classified the ground state of each dimer, in both the neutral and anionic state, as having a pronounced multi-reference character, as well as an effective bond order indicative of a quintuple bond.

Figure S4. The projection of the Ev[N,M]E_v[N,M] surface onto the N−MN-M plane for the CrMo (A), VCr (B) and VMo (C) hetero-dimers based on CASSCF/CASPT2 reference calculations for vertical charge and spin excitations [89]. The y-axis is labelled in terms of total charge as opposed to total electron count.

For each system, the total energy values are relative to the ground state total energy of the neutral dimer in its lowest energy magnetization state. Note that only a subset of the full Ev[N,M]E_v[N,M] surface is presented for these systems, this is due to the limited data available. If any additional reference data points not currently included are found to be low enough in energy, they could also affect the shape of the Ev[N,M]E_v[N,M] surfaces shown here.

SM-VIII. Model crystal/ligand field-dependent tilted plane structures for a d-orbital subspace

The energies of localized electronic states within a solid material also obey the tilted plane condition in the limit where the subspace-bath interaction energy varies linearly with spin resolved occupancy. This of course includes the case where the subspace-bath interaction becomes negligible. Figure S5 displays sample tilted plane structures for a dd-orbital subspace in the subspace-bath linear interaction limit under octahedral, tetrahedral, and square-planar crystal field splittings. The total energy of the electronic states with integer spin-resolved orbital occupancies were approximated using the model

Sample tilted plane structures for a d-orbital subspace

Figure S5. Sample tilted plane structures for a dd-orbital subspace in the subspace-bath linear interaction limit, under octahedral (A), tetrahedral (B) and square-planar (C) crystal field splittings. In the model energy (see Eq. (35)), parameters were chosen to show a sample of the the rich coordination-dependence of flat plane shapes. The interaction parameters UU and VV were set at 11 eV and 10 eV, respectively, and the orbital energy levels were chosen so that ∑iϵi=−80\sum_i \epsilon_i = -80 eV. The set of orbital energy levels {−24,−24,−24,−4,−4}\{-24, -24, -24, -4, -4\}, {−25,−25,−10,−10,−10}\{-25, -25, -10, -10, -10\} and {−25,−25,−18,−10,−2}\{-25, -25, -18, -10, -2\}, in eV, were used for the octahedral, tetrahedral and square-planar crystal fields, respectively. The total energy varies linearly across each plane but, for ease of visualization, each plane is colored based on the average of the energies at the vertices. In each example, the energies are reported relative to the plane with lowest average energy, whose energy was set to 1 eV to allow a logarithmic color scale.

Emod=∑iϵini+U∑ini↑ni↓+V2∑i≠jninj(35)E^{\mathrm{mod}} = \sum_i \epsilon_i n_i + U \sum_i n_i^{\uparrow} n_i^{\downarrow} + \frac{V}{2} \sum_{i\ne j} n_i n_j \tag*{(35)}

where UU and VV are the intra-orbital and inter-orbital interaction parameters, respectively. The values ϵi\epsilon_i and niσn_i^{\sigma} are the energy level and spin-σ\sigma electron count, respectively, of orbital ii. For each crystal field splitting, and hence degeneracy pattern, a wide variety of possible tilted plane structures exist. Total-energy approximations defined for non-integer spin-occupancies arise in a wide variety of DFT-related methods, e.g., in DFT+UU type XC-corrections and in machine learning proxies for total-energy components. These and perhaps next-generation XC approximations, as we approach exactitude, could beneficially (recalling Figure S5 (a) change expressions discretely between different tilted-plane segments, and (b) change in the manner of that change, depending on symmetry-imposed degeneracy patterns.

References

  1. [1]P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964).
  2. [2]W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
  3. [3]A. M. Teale, T. Helgaker, A. Savin, C. Adamo, B. Aradi, A. V. Arbuznikov, P. W. Ayers, E. Jan Baerends, V. Barone, P. Calaminici, E. Cances, E. A. Carter, P. Kumar Chattaraj, H. Chermette, I. Ciofini, T. Daniel Crawford, F. D. Proft, J. F. Dobson, C. Draxl, T. Frauenheim, E. Fromager, P. Fuentealba, L. Gagliardi, G. Galli, J. Gao, P. Geerlings, N. Gidopoulos, P. M. W. Gill, P. Gori-Giorgi, A. Görling, T. Gould, S. Grimme, O. Gritsenko, H. J. Aagaard Jensen, E. R. Johnson, R. O. Jones, M. Kaupp, A. M. Köster, L. Kronik, A. I. Krylov, S. Kvaal, A. Laestadius, M. Levy, M. Lewin, S. Liu, P.-F. Loos, N. T. Maitra, F. Neese, J. P. Perdew, K. Pernal, P. Pernot, P. Piecuch, E. Rebolini, L. Reining, P. Romaniello, A. Ruzsinszky, D. R. Salahub, M. Scheffler, P. Schwerdtfeger, V. N. Staroverov, J. Sun, E. Tellgren, D. J. Tozer, S. B. Trickey, C. A. Ullrich, A. Vela, G. Vignale, T. A. Wesolowski, X. Xu, and W. Yang, Phys. Chem. Chem. Phys. 24, 28700 (2022).
  4. [4]S. H. Vosko, L. Wilk, and M. Nusair, Can. J. Phys. 58, 1200 (1980).
  5. [5]J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).
  6. [6]A. D. Becke, Phys. Rev. A 38, 3098 (1988).
  7. [7]A. D. Becke, J. Chem. Phys. 98, 5648 (1993).
  8. [8]J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003).
  9. [9]C. Lee, W. Yang, and R. G. Parr, Phys. Rev. B 37, 785 (1988).
  10. [10]J. P. Perdew, K. Burke, and Y. Wang, Phys. Rev. B 54, 16533 (1996).
  11. [11]J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992).
  12. [12]J. Sun, A. Ruzsinszky, and J. P. Perdew, Phys. Rev. Lett. 115, 036402 (2015).
  13. [13]J. P. Perdew, R. G. Parr, M. Levy, and J. L. Balduz, Phys. Rev. Lett. 49, 1691 (1982).
  14. [14]W. Yang, Y. Zhang, and P. W. Ayers, Phys. Rev. Lett. 84, 5172 (2000).
  15. [15]A. J. Cohen, P. Mori-Sánchez, and W. Yang, J. Chem. Phys. 129, 121104 (2008).
  16. [16]P. W. Ayers, J. Math. Chem. 43, 285 (2008).
  17. [17]J. F. Janak, Phys. Rev. B 18, 7165 (1978).
  18. [18]T. Koopmans, Physica 1, 104 (1934).
  19. [19]J. P. Perdew and M. Levy, Phys. Rev. B 56, 16021 (1997).
  20. [20]L. Kronik and S. Kümmel, Phys. Chem. Chem. Phys. 22, 16467 (2020).
  21. [21]J. P. Perdew and M. Levy, Phys. Rev. Lett. 51, 1884 (1983).
  22. [22]L. J. Sham and M. Schlüter, Phys. Rev. Lett. 51, 1888 (1983).
  23. [23]E. Sagvolden and J. P. Perdew, Phys. Rev. A 77, 012517 (2008).
  24. [24]W. Yang, A. J. Cohen, and P. Mori-Sánchez, J. Chem. Phys. 136, 204111 (2012).
  25. [25]P. Mori-Sánchez and A. J. Cohen, Phys. Chem. Chem. Phys. 16, 14378 (2014).
  26. [26]T. Gould and J. Toulouse, Phys. Rev. A 90, 050502(R) (2014).
  27. [27]A. C. Burgess, E. Linscott, and D. D. O’Regan, J. Chem. Phys. 159, 211102 (2023).
  28. [28]G. K.-L. Chan, J. Chem. Phys. 110, 4710 (1999).
  29. [29]P. Mori-Sánchez, A. J. Cohen, and W. Yang, Phys. Rev. Lett. 102, 066403 (2009).
  30. [30]J. P. Perdew and E. Sagvolden, Can. J. Chem. 87, 1268 (2009).
  31. [31]X. De Vriendt, L. Lemmens, S. De Baerdemacker, P. Bultinck, and G. Acke, J. Chem. Theory Comput. 17, 6808 (2021).
  32. [32]X. De Vriendt, D. Van Hende, S. De Baerdemacker, P. Bultinck, and G. Acke, J. Chem. Phys. 156, 244115 (2022).
  33. [33]B. G. Janesko, Chem. Soc. Rev. 50, 8470 (2021).
  34. [34]T. Gál and P. Geerlings, J. Chem. Phys. 133, 144105 (2010).
  35. [35]A. M. Malek and R. Balawender, “Discontinuities of energy derivatives in spin-density functional theory,” (2013), arXiv:1310.6918 [physics].arxiv.org/abs/1310.6918
  36. [36]X. D. Yang, A. H. G. Patel, R. A. Miranda-Quintana, F. Heidar-Zadeh, C. E. González-Espinoza, and P. W. Ayers, J. Chem. Phys. 145, 031102 (2016).
  37. [37]R. Cuevas-Saavedra, D. Chakraborty, S. Rabi, C. Cárdenas, and P. W. Ayers, J. Chem. Theory Comput. 8, 4081 (2012).
  38. [38]J. P. Perdew, Int. J. Quantum Chem. 28, 497 (1985).
  39. [39]P. Borlido, T. Aull, A. W. Huran, F. Tran, M. A. L. Marques, and S. Bottl, J. Chem. Theory Comput. 15, 5069 (2019).
  40. [40]A. J. Cohen, P. Mori-Sánchez, and W. Yang, Phys. Rev. B 77, 115123 (2008).
  41. [41]A. Ruzsinszky, J. P. Perdew, G. I. Csonka, O. A. Vydrov, and G. E. Scuseria, J. Chem. Phys. 125, 194112 (2006).

Paper details

vlnc:W_gng39cc6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents