Abstract
Stationarity of electron probability distribution within the resolution of atomic orbitals is considered involving some concepts from Orbital Communication Theory and the theory of Markov Processes. A new method of evaluating electron conditional probabilities based on natural orbitals is proposed and briefly discussed.
Introduction
In the last halfcentury the electron population analyses (EPA) turned out to be very useful and commonly used tools in probing the electronic structure and chemical reactivity of molecules. In brief, EPAprocedures give rise to partition the electron density of the whole molecule between atoms, chemical bonds, molecular fragments, etc. Such partition can be performed within either physical or Hilbert space of molecular orbitals (MO) and there is a multitude of diverse EPAprocedures among which the most commonly used are those proposed by: Mulliken [1, 2], Löwdin [3, 4], Weinhold [5], Bader [6], Hirshfeld [7] and MerzKollman [8].
Although this is a field of research that seems to be regarded by some scientists as exhausted, in this short paper we would like to address some of the general issues respecting stationarity and uniqueness of electron probability distributions within the framework of MO theory and resolution of atomic orbitals (AO). Our considerations will be confined only to closedshell ground states at two levels of theory, HartreeFock (HF) and configuration interaction (CI) [9]. The main purpose of this article is to briefly introduce a new method of determining stationary electron probability distributions, based on natural orbitals (NO) and involving some concepts from the Orbital Communication Theory (OCT) of the chemical bond [10–12] and the theory of Markov Processes [13].
Stationarity from idempotency
For simplicity and transparency of definitions we assume without loss of generality the wavefunctions to be real. First, let us consider the onedeterminant wavefunction of the groundstate molecular system, \(\Psi _{HF}\rangle \), with \(N\) electrons doubly occupying the \(n^o\) lowest of MOs, \(\varvec{\varPhi }^o\rangle \), from the Hilbert space of the all \(n\) orthonormal (spinless) molecular orbitals,
generated as the linear combinations of basis functions \(\varvec{\chi }\rangle \), representing orthogonal atomic orbitals,
where square matrix C groups the relevant LCAO MO expansion coefficients,
Obviously, \(n=n^o+n^v\), and superscripts \(o\) and \(v\) refer to occupied and virtual MOsubspaces, respectively.
A set of such MOs uniquely defines the system oneelectron density,
where \({\gamma }\) is the so called chargeandbondorder (CBO) matrix,
satisfying the following idempotency relation appropriate for the closedshell systems (duodempotency relation):
Diagonal elements of symmetric matrix CBO, \(\{\gamma _{\mu ,\mu }\}\), can be regarded as effective electron occupations of AOs, \(N_{\mu }\), determining the corresponding electron probability in AO resolution:
with the appropriate normalization condition
The second equality in (6) allows one to define the so called conditional probability matrix P
and the following normalization condition
According to OCT, the element of CBOmatrix, \(\gamma _{\nu ,\mu }\), represents the appropriately renormalized amplitude of the corresponding conditional probability \(P_{\nu \mu }\) of ”observing” in atomic orbital \(\chi _\nu \rangle \) the electron originally assigned to \(\chi _\mu \rangle \). Consequently, \(P_{\nu \mu }>0\) gives rise to direct communication (throughspace interaction [14, 15]) between AOs, \(\mu \rightarrow \nu \), while a sequence of \(r\) seriallyconnected communications determined by conditional probabilities \(P_{\nu \mu }^r\),
refers to indirect AOcommunication (throughbridge interaction [14–16]). It has to be noticed that such informationpropagation cascade of AOchannels one can recognize as an analogue to the timehomogeneous Markov chain with a finite state space [13]; then \(P_{\nu \mu }^r\) represents the relevant \(r\)step transition probability.
Due to idempotency relation (6) the electron distribution p (row vector) is said to be stationary in the sense of Markov if:
As stated by the PerronFrobenius theorem [17], for any arbitrary matrix with nonnegative elements of each column normalized to 1 there is always at least one stationary probability distribution. In fact, as follows from the second equality of the last equation, the stationary distribution can be ragarded as a normalized left eigenvector of the conditional probability matrix associated with the eigenvalue of 1. Therefore, for given matrix P one can find the relevant vector p by solving the eigenproblem (12) but, in general, the resulting probability distribution is not guaranteed to be unique, i.e. there can be more that one eigenvector of P associated with the eigenvalue 1. It follows from the PerronFrobenius theorem that the uniqueness of the stationary electron distribution is always assured if the \(x\)th power of the conditional probability matrix converges to a rankone matrix in which each row is the stationary distribution in question,
where 1 stands for the column vector with all entries equal 1. Essentially, for a given conditional probability matrix its stationary distribution p is nondegenerated provided that:

matrix P represents an aperiodic Markov chain which virtually means that for all atomic orbitals \(\{\chi _\mu \rangle \}\) the direct selfcommunication \(\mu \rightarrow \mu \) is always possible, \(P(\mu \mu )>0\). In practice, aperiodicity can be always attained by appropriate reduction of the AOspace size, i.e. by removing atomic orbitals for which \(\gamma _{\mu ,\mu }<\epsilon \), where \(\epsilon \) represents the fixed threshold value;

matrix P represents an irreducible Markov chain. i.e. if all atomic orbitals are directly and/or indirectly communicated with each other. Also this requirement is straightforward to achieve by eliminating (if present) those AOs for which \(\gamma _{\mu \mu }>2\epsilon \) (ussually core orbitals). The subspace od mutually communicated atomic orbitals defines a single communicating class.
Conditional probability (9) and relation (12) are crucial for the electron population analysis. Indeed, each row of matrix P allows one to decompose effective electron population \(N_{\mu }\) into terms related to localized (diagonal) and emphdelocalized (offdiagonal) populations, i.e. \(N_{\mu ,\mu }\) and \(N_{\mu ,\nu }\), respectively:
In atomic resolution the electron population on atom X can be decomposed as follows:
Here, \(N_{X,X}\) stands for the number of electrons well localized on atom X (noninteracting “chemically” with electrons from other atoms, e.g. lone pairs, core electrons) and \(V_X\) is the atomiccovalency index counting up the population of electrons delocalized over chemical bonds with all other atoms,
\(W_{X,Y}\) stands for the standard (quadratic) bondcovalency index originally proposed by Wiberg [18]. However, one should realize that electron populations and covalencies from Eqs. (15)–(16) calculated within representation of orthogonal AOs assume reliable and chemically meaningful values only when minimal set of conventionally atomassigned atomic orbitals is used (this problem has been recently reexamined within the framework of OCTapproach in [19]). Such minimal basis set of orthogonal atomic orbitals one can always obtain using various methods, e.g. [20, 21].
Stationarity without idempotency
Unfortunately, Definitions (5)–(10) are correct only for onedeterminant wavefunctions of closedshell systems. In typical CItype calculations the electron density matrix \(\mathbf{\Gamma }\) of the multideterminant wavefunction represented by \(\varPsi _{CI}\rangle \) is commonly expressed by means of its eigenvectors—natural orbitals \(\varvec{\varTheta }\rangle \), and the corresponding eigenvalues—occupation numbers \(\varvec{\uplambda },\)
where the electron density operator \(\hat{\varGamma }\) is defined as follows:
Then
with normalization \(\text{ Tr}\varvec{\varGamma }=N\) and the electron probability distribution p is determined by diagonal elements \(\{\varGamma _{\mu ,\mu }\}\)
However, since the electron density operator \(\hat{\varGamma }\) is generally not idempotent,
normalization condition (10) cannot be fulfilled and hence offdiagonal elements of the relevant density matrix, \(\{ \varGamma _{\mu ,\nu } \}\) cannot be further regarded as amplitudes of corresponding conditional probabilities. In other words, for the hypothetical conditional probability matrix \(\mathbf{P}^\prime \) obtained from appropriate renormalization of each row of matrix of elements \(\{\varGamma _{\mu ,\nu }^2\}\) its stationary distribution \(\mathbf{p}^\prime \) (left eigenvector associated with the eigenvalue 1) differs from p and
To solve this problem we propose in this paper the following alternative definition of conditional probability \(P_{\nu \mu }\), by means of natural orbitals and their occupation numbers:
where
Therefore, one can interpret \(B^i_{\mu ,\nu }\) as the contribution of \(i\)th NO to the corresponding electron population \(N_{\mu ,\nu }\) provided that all orbitals \(\{\varTheta _j\rangle , j=1,\cdots ,i1 \}\) are fully occupied.
In the onedeterminant case of closedshell molecular system all \(n^o\) occupied natural orbitals correspond to degenerated eigenvalue \(\uplambda _i=2\) and therefore can be easily separated from the (usually larger) subspace of \(n^v\) unoccupied orbitals (\(\uplambda _i=0\)). Due to this separation and degeneracy of \(\{\uplambda _i\}\) it can be proved that both definitions of conditional probability \(P_{\nu \mu }\), (9) and (23), are fully equivalent, regardless of orbital order:
For onedeterminant wavefuntions the atomic charges and Wibergtype bond covalencies (14)–(16) based on the density matrix (5) are commonly known to be invariant with respect to unitary transformation of molecular orbitals. Thus, from the equivalence of definitions (9) and (23) one can deduce the same transformational properties of conditional probabilities (23).
It is also straightforward to prove that, for the newly proposed definition (23), stationarity of probability distribution p is preserved for any arbitrary permutation of eigenvectors and eigenvalues of density matrix. First, let us sum up elements \(B_{\mu ,\nu }^i\) over all atomic orbitals,
The last equality in the preceding equation results from orthonormality of natural orbitals, \(\langle \varTheta _i\varTheta _j\rangle =\delta _{i,j}\). Now, using the result from (26) we can easily prove the appropriate normalization requirement,
Since electron probabilities (and the corresponding populations) are invariant with respect to interchange of any pair of natural orbitals there are formally \(r_P=n!\) different matrices P satisfying (12) for nondegenerated \(\varvec{\uplambda }\) (i.e. \(g_s=1\) for \(s=1,\cdots ,n\)). However, if degeneracy occurs (or certain eigenvalues below a fixed threshold can be neglected) and there are only \(n^\prime \) different eigenvalues (i.e. \(g_s\ge 1\) for \(s=1,\cdots ,n^\prime \)) then the number of possible conditional probability matrices reduces to
where \(g_s\) stands for the degeneracy of \(s\)th possible eigenvalue.
In the case of onedeterminant calculations subspaces of occupied and virtual orbitals are strictly separated and there are exactly \(n^o!\) permutations of \(\varvec{\uplambda }\) that lead to the same conditional probability matrix. Furthermore, due to orthonormality of natural orbitals and basis functions, the maximal separation of these subspaces ensures that all \(P_{\nu \mu }\ge 0\). All permutations of “mixed” occupied and unoccupied NOs can lead to negative values of \(P_{\nu \mu }\); this is mainly because of predominatingly antibonding character of virtual subspace with relation to the occupied one. This facts allow one to draw a conclusion that for the multideterminant calculations the natural generalization of the maximal separation requirement is the following condition:
Thus, for groundstate systems the conditional probability matrix P obtained from permutations of NOs for which \(\uplambda _1\le \cdots \le \uplambda _{i}\le \uplambda _{i+1}\le \cdots \le \uplambda _n\), reveals the minimal dependence on order of NOs within the particular (conventionally separated) subspace, high occupied (\(\uplambda _i\approx 2\)) or low occupied orbitals (\(\uplambda _i\approx 0\)).
Summary
It was the main goal of this short paper to demonstrate that idempotency property of the density operator is not an indispensable condition for the stationarity of electron probability distribution in molecular systems. We have introduced a simple method of evaluation of conditional probabilities for the multideterminant wavefunction. Moreover, the newly proposed definition based on natural orbitals successfully displaces hitherto prevailing one for the onedeterminant wavefunctions. Since presented methodology introduces some arbitrariness due to orbital order, it has been argued to use the criterion of maximal separation of nearlydegenerated orbital subspaces in groundstate calculations.
The proposed method can be also extended to cover openshell molecular systems as well as the excited states. However, this requires a more insightful investigation and will be the subject of the seperate article.
References
 1.
R.S. Mulliken, J. Chem. Phys. 23, 1833 (1955)
 2.
R.S. Mulliken, J. Chem. Phys. 23, 1841 (1955)
 3.
P.O. Löwdin, J. Chem. Phys. 18, 365 (1950)
 4.
D. Szczepanik, J. Mrozek, Chem. Phys. Lett. 521, 157 (2012)
 5.
A.E. Reed, R.B. Weinstock, F. Weinhold, J. Chem. Phys. 83, 735 (1985)
 6.
R.F.W. Bader, P.M. Beddall, P.E. Cade, J. Am. Chem. Soc. 93, 3095 (1971)
 7.
F.L. Hirshfeld, Theor. Chim. Acta. 44, 129 (1977)
 8.
U.C. Singh, P.A. Koolman, J. Comput. Chem. 5, 129 (1984)
 9.
F. Jensen, Introduction to Computational Chemistry, 2nd edn. (Wiley, West Sussex, 2007)
 10.
R.F. Nalewajski, Information Theory of Molecular Systems (Elsevier, Amsterdam, 2006)
 11.
R.F. Nalewajski, Information Origins of The Chemical Bond (Nova Science Publishers, Hauppauge, 2010)
 12.
R.F. Nalewajski, D. Szczepanik, J. Mrozek, Adv. Quantum Chem. 61, 1 (2011)
 13.
A.A. Markov, Sci. Context 19(4), 591 (2006)
 14.
R.F. Nalewajski, J. Math. Chem. 49, 371 (2011)
 15.
R.F. Nalewajski, Int. J. Quantum Chem. 112, 2355 (2012)
 16.
R.F. Nalewajski, J. Math. Chem. 49, 546 (2011)
 17.
J. Keener, SIAM Rev. 35, 80 (1993)
 18.
K.B. Wiberg, Tetrahedron 24, 1083–1096 (1968)
 19.
R.F. Nalewajski, D. Szczepanik, J. Mrozek, J. Math. Chem. 50, 1437 (2012)
 20.
D. Szczepanik, J. Mrozek, Comput. Theor. Chem. 996, 103 (2012)
 21.
W.C. Lu, C.Z. Wang, M.W. Schmidt, L. Bytautas, K.M. Ho, K. Ruedenberg, J. Chem. Phys. 120, 2629 (2004)
Author information
Affiliations
Corresponding author
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.
About this article
Cite this article
Szczepanik, D., Mrozek, J. Stationarity of electron distribution in groundstate molecular systems. J Math Chem 51, 1388–1396 (2013). https://doi.org/10.1007/s1091001301538
Received:
Accepted:
Published:
Issue Date:
Keywords
 Stationary distribution
 Markov chain
 Density matrix
 Conditional probability
 Natural orbital