Natural Orbital Occupation Numbers

The familiar high-school chemistry picture of electrons neatly filling orbitals with integer counts—0, 1, or 2—provides a useful but incomplete foundation for understanding molecular structure. This simple framework, rooted in the Hartree-Fock approximation, gives us a language for chemical bonding but overlooks a crucial aspect of reality. Electrons are not independent; they actively repel and avoid one another in a complex, high-speed dance known as electron correlation. This phenomenon is invisible to simple models but is fundamental to understanding chemical reactivity, bond breaking, and molecular properties. To move beyond approximation and grasp the true electronic structure, we need a more sophisticated accounting system.

This article introduces natural orbital occupation numbers (NONs) as the precise language for describing electron correlation. These numbers provide a direct, quantitative measure of the average electron count in a set of unique "natural" orbitals, revealing the subtle details of the electronic dance. In the following chapters, we will explore how these non-integer values arise and what they tell us about the different flavors of correlation. We will then see these concepts in action, demonstrating how NONs serve as a powerful diagnostic tool for chemists, a blueprint for building better theories, and a bridge between different schools of chemical thought.

In our school-day picture of chemistry, electrons are well-behaved tenants in an atomic or molecular high-rise. They fill up the orbital "apartments" starting from the ground floor, two to a room, one spinning "up" and the other "down". An orbital is either full (with 2 electrons), half-full (with 1), or empty (with 0). The world, in this view, is governed by simple integers. This wonderfully straightforward model, known in the trade as the ​​Hartree-Fock approximation​​, is the bedrock of much of our chemical intuition. It builds molecules, explains bonds, and gives us a language to speak about electronic structure. But, as is so often the case in physics, the simple picture is just the first, beautiful draft of a much richer and more subtle story.

The electrons in a molecule are not independent tenants; they are a deeply interconnected family. They are all negatively charged, so they constantly repel one another. They engage in an intricate, high-speed dance to stay out of each other's way. This complex choreography is what physicists and chemists call ​​electron correlation​​. The simple integer-based model neglects this dance; it treats each electron as moving in an average field created by all the others. To truly understand the behavior of molecules—why some bonds break easily, why some molecules are colored, and how chemical reactions really happen—we must account for this correlation. And to do that, we need a new way of counting.

Imagine you could take a snapshot of all the electrons in a molecule at a given instant. You'd find them in some configuration. Take another snapshot, and they'll be somewhere else. If you could average over an immense number of these snapshots, you could ask, "On average, how many electrons are we likely to find in the region of space we call orbital $\phi_1$?" The simple model says the answer must be 2 or 1 or 0. The real world says otherwise.

The proper quantum mechanical tool for this job is the ​​one-particle reduced density matrix​​, or ​​1-RDM​​. It might sound intimidating, but its role is simple: it's the ultimate bookkeeper of the electrons' average distribution. When we ask the 1-RDM to organize itself into its most natural representation, it gives us a special set of orbitals called ​​natural orbitals​​, and for each one, it provides a number—the ​​natural orbital occupation number (NON)​​. These numbers are the answer to our question. They are the true, average occupancy of each natural orbital.

These NONs, let's call them $n_i$ for the $i$-th orbital, follow a few fundamental rules derived directly from the laws of quantum mechanics for fermions:

1. ​​The Bounds:​​ For a spatial orbital that can hold up to two electrons (one spin-up, one spin-down), the occupation number $n_i$ must lie in the range $0 \le n_i \le 2$. It can be any real number in between! If we were looking at spin-orbitals (which can only hold one electron), the bounds would be $0 \le n_i \le 1$.

2. ​​The Sum Rule:​​ If you add up the occupation numbers of all the natural orbitals in a system, you must get the total number of electrons, $N$. That is, $\sum_i n_i = N$. The bookkeeping is exact.

3. ​​The Integer Limit:​​ When do we get our simple integer picture back? This happens only when the true, correlated wavefunction of the system is perfectly described by a single Slater determinant—the mathematical object that represents the simple "orbitals as boxes" picture. In this limiting case, and only in this case, the occupation numbers are all exactly 2 or 0 (for a closed-shell system).

Any deviation from these integer values is a direct measure of electron correlation. The simple model is no longer just an approximation; the NONs tell us precisely how much and in what way it deviates from reality.

How do these fractional numbers arise? Let's peek under the hood with a simple, hypothetical two-electron atom modeled with just two available spatial orbitals, $\phi_1$ and $\phi_2$. In the simplest picture, the ground state has both electrons in the lowest energy orbital, $\phi_1$. We would write this configuration as $(\phi_1)^2$. The occupation numbers would be $n_1=2$ and $n_2=0$.

But what if the electrons can lower their overall energy by occasionally venturing into the higher orbital, $\phi_2$? The true state might be a mixture, a ​​superposition​​, of the $(\phi_1)^2$ configuration and the doubly-excited $(\phi_2)^2$ configuration. This is the central idea of the ​​Configuration Interaction (CI)​​ method. The wavefunction is no longer just one configuration, but a weighted sum:

$\Psi = c_1 \Phi_1 + c_2 \Phi_2$

Here, $\Phi_1$ represents the $(\phi_1)^2$ configuration and $\Phi_2$ represents $(\phi_2)^2$. The coefficients $c_1$ and $c_2$ tell us how much each configuration contributes. Because the total probability must be one, we know that $|c_1|^2 + |c_2|^2 = 1$.

When you perform the bookkeeping for this mixed state, a beautiful result emerges. The natural orbitals are still $\phi_1$ and $\phi_2$, but their occupation numbers are now:

$n_1 = 2|c_1|^2 \quad \text{and} \quad n_2 = 2|c_2|^2$

Notice that $n_1 + n_2 = 2(|c_1|^2 + |c_2|^2) = 2(1) = 2$. The sum rule holds perfectly! Suppose a calculation finds that the ground state is described by $\Psi = \frac{3}{\sqrt{10}} \Phi_1 - \frac{1}{\sqrt{10}} \Phi_2$. Then the occupations are $n_1 = 2 \left| \frac{3}{\sqrt{10}} \right|^2 = 2(\frac{9}{10}) = 1.8$ and $n_2 = 2 \left| -\frac{1}{\sqrt{10}} \right|^2 = 2(\frac{1}{10}) = 0.2$. The electrons now spend, on average, 90% of their time as a pair in $\phi_1$ and 10% of their time as a pair in $\phi_2$. A total of $0.2$ of an electron's worth of occupancy has been "promoted" from the lower orbital to the higher one to better describe the correlated motion. This is the mechanism in its simplest form.

The fact that occupations can be fractional is interesting, but their true power comes from using them as a diagnostic tool. The values of the NONs can tell us not just that correlation is present, but what kind of correlation it is. Chemists generally speak of two main flavors.

​​Dynamic correlation​​ is the ubiquitous, moment-to-moment avoidance of electrons. It's like the subtle dance people do in a crowded elevator to maintain their personal space. This kind of correlation involves a vast number of configurations, each making a tiny contribution to the true wavefunction. Its signature in the NONs is therefore subtle: the occupations of orbitals that would be full in the simple model are slightly depleted (e.g., 1.99, 1.98), and a huge number of "virtual" orbitals that would be empty gain tiny populations (e.g., 0.01, 0.005).

A perfect example is the helium dimer, $\text{He}_2$. The two helium atoms are closed-shell systems, quite happy on their own. They are bound together only by the very weak and subtle van der Waals forces, which are a direct manifestation of dynamic correlation. A high-level calculation on $\text{He}_2$ shows exactly this pattern: two orbitals with occupations very near 2, and the other two with occupations very near 0. The single-determinant picture is qualitatively correct, and the small deviations quantify the weak, dynamic correlation that gives rise to the bond.

​​Static (or Nondynamical) correlation​​ is a much more dramatic phenomenon. It occurs when the system finds itself in a situation where two or more electronic configurations have very similar energies. Here, the simple picture of one dominant configuration isn't just slightly wrong; it's qualitatively wrong. The system is fundamentally a mixture of these configurations.

The smoking gun for strong static correlation is natural orbital occupations that are ​​far from 0 or 2​​, especially occupations close to 1.

The canonical example is the breaking of a chemical bond, like in the hydrogen molecule, $\text{H}_2$. At its normal bond length, $\text{H}_2$ is well-described by placing both electrons in the bonding $\sigma_g$ orbital. The NONs are roughly $n_g \approx 1.98$, $n_u \approx 0.02$. But as you pull the two hydrogen atoms apart, the bonding ($\sigma_g$) and antibonding ($\sigma_u$) orbitals become closer and closer in energy. At infinite separation, the system is just two separate hydrogen atoms, each with one electron. To describe this correctly, the wavefunction must be an equal mixture of the $(\sigma_g)^2$ and $(\sigma_u)^2$ configurations. This forces the occupation numbers to become $n_g = 1$ and $n_u = 1$. Each orbital contains exactly one electron on average. The journey of the NONs from $(2, 0)$ to $(1, 1)$ as the bond stretches is the quintessential signature of the onset of strong static correlation.

This isn't just a story about H$_2$. For any molecule whose bond is stretched, like F$_2$, we see the same pattern: occupations that are close to (2, 0) at equilibrium move towards (1, 1) upon dissociation. Looking at a set of calculated occupation numbers, such as $\{1.995, 1.987, 1.152, 0.848, 0.013, 0.005\}$, a trained eye can immediately diagnose the situation. The first two orbitals are strongly occupied, the last two are nearly empty, but the middle two, with occupations near 1, scream "strong static correlation!" This molecule has what chemists call a strong ​​multireference character​​.

Our eyes are good at spotting patterns, but science delights in quantification. Can we distill the complex list of NONs into a single number that measures the degree of static correlation? Several such measures exist.

One intuitive idea is to count the "number of effectively unpaired electrons". For a perfectly closed-shell system, this number is 0. For a system with two truly unpaired electrons (a diradical), this number should be 2. A clever formula does just this:

$N_u = \sum_i n_i(2-n_i)$

Let's test it. If $n_i=2$ (fully paired) or $n_i=0$ (empty), the term $n_i(2-n_i)$ is zero. So, for a perfect single-determinant state, $N_u=0$. What if we have two orbitals with occupations $n=1$, as in stretched H$_2$? Then $N_u = 1(2-1) + 1(2-1) = 2$. It works! For the system with NONs $\{1.98, 1.90, 1.08, 0.98, 0.04, 0.02\}$, a quick calculation gives $N_u \approx 2.34$, confirming the presence of about two effectively unpaired electrons and significant static correlation.

Another beautiful connection is to information theory. A state with integer occupations {2, 0, 0, ...} is a state of perfect information—we know exactly which orbitals are filled. A state with fractional occupations {1, 1, 0, ...} reflects uncertainty; the electrons are shared between two orbitals. The ​​Shannon entropy​​ of the normalized NON distribution ($p_i=n_i/N$) quantifies this uncertainty. As H$_2$ dissociates, the entropy of its NONs increases from 0 to its maximum value of $\ln(2)$, beautifully charting the breakdown of the single-reference picture.

So far, we have been discussing the properties of the true wavefunction, or at least a very sophisticated approximation to it. But what about the workhorse methods that many chemists use? The ​​Unrestricted Hartree-Fock (UHF)​​ method is a popular single-determinant approach that tries to handle bond-breaking by allowing the spin-up and spin-down electrons to occupy different spatial orbitals.

This often gives a better energy, but at a cost: the resulting wavefunction is no longer a pure spin state (e.g., a "singlet" becomes contaminated with triplet character). This is called ​​spin contamination​​. Now for a very subtle, but crucial point. Since a UHF wavefunction is still a single Slater determinant, the occupations of its fundamental ​​natural spin-orbitals​​ are still exactly 1 and 0. There is no contradiction.

So where do the fractional occupations often associated with UHF come from? They arise when one takes the spin-contaminated UHF determinant and calculates the occupations of the spatial orbitals (by summing the up- and down-spin contributions). These fractional occupations are an artifact of the spin-symmetry breaking. They are a clue that the single-determinant method is struggling, but they are not the same thing as the genuine fractional occupations of a true, multiconfigurational, a spin-pure wavefunction. Distinguishing between a physical effect (static correlation) and a methodological artifact (spin contamination) is one of the fine arts of the trade, a reminder that we must always question the assumptions behind our models.

The world of electrons is not one of simple integers. It is a world of fractional occupancies, of subtle dynamic dances and dramatic static rearrangements. The natural orbital occupation numbers are our window into this world, providing a language that is both quantitative and rich with physical intuition, turning the abstract problem of electron correlation into a story we can read and understand.

Now that we have explored the principles behind natural orbitals and their occupation numbers, we arrive at the most exciting part of our journey. How do these abstract mathematical concepts actually help us understand the world? It turns out that this set of numbers, the eigenvalues of the one-particle density matrix, is something of a quantum Rosetta Stone. It allows us to translate the forbiddingly complex, high-dimensional dance of electrons within a molecule into a language that is not only comprehensible but also profoundly insightful and practical. In this chapter, we will see these numbers in action, serving as a chemist's diagnostic tool, an architect's blueprint for building better theories, and even a universal translator between different schools of chemical thought.

Imagine a physician trying to understand a patient's health. A simple set of numbers—temperature, blood pressure, heart rate—can reveal a great deal about the body's internal state. Natural orbital occupation numbers (NONs) serve a similar role for molecules. They are vital signs that allow us to diagnose the health and character of chemical bonds.

For a simple, well-behaved chemical bond, like the double bond in an ethylene molecule in its stable, planar form, the story is straightforward. This is a system that the simple Aufbau principle, learned in introductory chemistry, describes rather well. We expect one bonding orbital to be fully occupied by two electrons and the corresponding antibonding orbital to be empty. And indeed, a detailed calculation reveals NONs that are very close to the integers we expect: the bonding $\pi$ orbital has an occupation near $2.0$, while the antibonding $\pi^*$ orbital has an occupation near $0.0$. The slight deviation from perfect integers, for instance, values like $1.98$ and $0.02$, is the subtle whisper of what we call dynamic correlation—the ever-present, jittery dance of electrons as they try to avoid one another. The system is, for all intents and purposes, "healthy" and well-described by a single electronic configuration.

But what happens when we stress the system? Let's take our ethylene molecule and begin to twist it around its central carbon-carbon bond. As the geometry deforms, the overlap that creates the $\pi$ bond is destroyed. At a $90^\circ$ twist, the bond is effectively broken. What do our quantum vital signs tell us now? The NONs undergo a dramatic transformation. The occupation of the bonding $\pi$ orbital plummets from nearly $2$ down to almost exactly $1.0$, while the occupation of the antibonding $\pi^*$ orbital climbs from nearly $0$ up to almost exactly $1.0$.

This is a clear and unambiguous signal that something fundamental has changed. The system can no longer be described by a single, simple configuration. The occupations of $1.0$ and $1.0$ are the definitive signature of a diradical: a molecule with two electrons that have "uncoupled" and now occupy two different orbitals. The wavefunction has become an equal mixture of at least two competing electronic configurations, a classic case of strong static correlation. The simple Aufbau principle has completely broken down, and the NONs have provided us with a clear, quantitative diagnosis of the bond's failure.

Not all molecules with fractional occupations are "sick," however. Some are just naturally complex. Consider the ozone molecule, $O_3$, famous for the resonance structures chemists draw to represent its bonding. A quantum chemical calculation reveals active orbitals with occupations that are far from any integer, for example, $1.75$, $1.00$, and $0.25$. This isn't a sign of a broken bond, but rather the quantum mechanical reality of resonance. The true state of the molecule is a superposition, a weighted average of multiple electronic arrangements. The NONs are the direct expression of this weighted average, giving us a much more precise and quantitative picture of resonance than the simple cartoons with double-ended arrows can provide.

This diagnostic tool is remarkably subtle. For the allyl radical ($C_3H_5$), another system with resonance, one might expect a similarly complex picture. Yet, a calculation reveals NONs for its $\pi$ system that are very close to integers: $1.94$, $1.02$, and $0.04$. This tells us something surprising: despite its resonance, the allyl radical is electronically quite simple, dominated by a single configuration where one orbital is doubly occupied, one is singly occupied, and one is empty. The resonance here is captured not by mixing configurations, but by the delocalized shape of the singly occupied orbital itself. The NONs, in this case, save us from over-interpreting the situation and applying a needlessly complex model.

Knowing the diagnosis is one thing; acting on it is another. Beyond their diagnostic power, NONs serve as an essential, practical tool for the computational chemist—an architect's blueprint for constructing accurate and efficient theoretical models.

The challenge in modern quantum chemistry is often one of complexity. The exact equations governing a molecule's electrons are too difficult to solve for all but the simplest systems. We must approximate. One of the most powerful strategies is to divide and conquer: we treat the few "problematic" electrons involved in strong static correlation with a very high-level, complex method (like in a CASSCF calculation), while treating the many well-behaved electrons with a simpler approach. The crucial first step is identifying these "problematic" orbitals. This is called choosing the active space.

How is this choice made? It could be a dark art, but NONs turn it into a science. A common and highly effective procedure is to perform a quick, less expensive preliminary calculation. We then compute the NONs from this initial guess. Experience has taught us that orbitals whose occupations deviate significantly from $0$ or $2$ are the ones involved in static correlation. So, we establish a threshold: any orbital whose occupation number $n_i$ falls within a certain range, for example $0.02 \lt n_i \lt 1.98$, is flagged for inclusion in the active space.

The choice of this threshold isn't arbitrary. It's rooted in the physics of electron correlation. The small deviations (occupations like $1.99$ or $0.01$) are typically due to the ubiquitous dynamic correlation. The large deviations (occupations like $1.8$ or $0.2$) are the signal of the strong static correlation we must treat with our most powerful methods. The threshold is our filter, allowing us to focus our computational firepower where it is most needed.

The utility doesn't end there. After performing a sophisticated CASSCF calculation using our chosen active space, we can again turn to NONs for quality control. Has our active space successfully captured all the strong correlation? Or has some of it "leaked" out into the orbitals we decided to treat simply? We can construct a diagnostic—a single number calculated from the final NONs—that quantifies the fraction of static correlation that exists outside our active space. If this number is small, we can be confident in our model and proceed with even more advanced calculations built upon it. If it's large, it's a red flag telling us we need to revise our blueprint and choose a larger active space. This is the self-correcting nature of computational science at its finest.

Perhaps the most beautiful application of natural orbitals is their ability to act as a unifying framework, a common language connecting seemingly disparate theoretical models of chemical bonding. For generations, chemists have been taught two foundational, yet rival, pictures: Molecular Orbital (MO) theory and Valence Bond (VB) theory.

MO theory describes electrons as belonging to the molecule as a whole, occupying delocalized orbitals that can span the entire system. VB theory, in contrast, tells a more local story, envisioning bonds as arising from the interaction of electrons in their original atomic orbitals, with the overall state described as a mixture of covalent and ionic structures. These two "languages" often seem to be telling completely different stories.

This is where natural orbitals step in as the universal translator. It turns out that any two-electron wavefunction, regardless of whether it was constructed from an MO or a VB starting point, can be re-expressed in an identical, simple, and unique "natural" form: a sum of configurations built from its own natural orbitals.

Consider the Lithium Hydride (LiH) molecule. We can write a VB wavefunction that is a mixture of a covalent (Li–H) and an ionic (Li$^+$H$^-$) structure. This looks very different from an MO description. However, if we take this VB wavefunction and compute its natural orbitals and their occupation numbers, we find a result that is directly comparable to a multi-configurational MO calculation. The occupations, say $n_1$ and $n_2$, give us the definitive, theory-independent measure of the electron correlation in the bond. The NONs reveal the fundamental physical reality that both theories are striving to describe. They show that MO and VB theory are not so much right or wrong as they are different perspectives on the same underlying quantum truth, a beautiful example of the unity of our scientific description of nature.

From a practical diagnostic tool to a guide for building robust theories and a bridge between competing ideas, natural orbital occupation numbers provide a powerful lens through which to view the intricate world of electrons in molecules. They are a testament to how an elegant mathematical concept can illuminate the path to a deeper and more unified understanding of chemistry.