SCAN Functional

In the quest to understand and predict the behavior of matter at the quantum level, scientists rely on Density Functional Theory (DFT), a framework that promises to calculate the properties of any system from its electron density alone. The key to this promise lies in the elusive exchange-correlation functional, a "universal recipe" that remains one of physics' great unsolved problems. Over the years, approximations have been organized into a hierarchy known as "Jacob's Ladder," with each rung offering a more refined view of the electronic world. However, lower-level approximations like the Generalized Gradient Approximation (GGA) suffer from a critical blindness, often failing to distinguish between fundamentally different types of chemical bonds.

This article delves into the Strongly Constrained and Appropriately Normed (SCAN) functional, a major advancement to the third rung of this ladder that addresses these shortcomings. By exploring SCAN, we will uncover a powerful design philosophy rooted in adhering to the fundamental rules of quantum mechanics rather than empirical fitting. In the following chapters, you will learn about the intricate principles and mechanisms that grant SCAN its discerning power, and you will journey through its vast applications across chemistry and materials science, revealing how this physics-based approach unlocks new frontiers in scientific research.

Imagine you are a master chef trying to create a universal recipe. A recipe that, given just one basic ingredient, can produce any dish imaginable, from a simple soup to the most elaborate feast. In the world of quantum mechanics, physicists and chemists are on a similar quest. Their "basic ingredient" is the electron density, $n(\mathbf{r})$, a simple function that tells us the probability of finding an electron at any point $\mathbf{r}$ in a molecule or material. Their "universal recipe" is a mathematical object called the ​​exchange-correlation functional​​, $E_{\mathrm{xc}}[n]$. If they had the exact form of this functional, they could, in principle, calculate the properties of any atom, molecule, or solid just from its electron density.

This grand recipe, however, remains one of the greatest unsolved problems in physics. What we have instead is a collection of increasingly sophisticated approximations, a conceptual ladder of progress that physicists have been climbing for decades. The Strongly Constrained and Appropriately Normed (SCAN) functional represents a major leap up this ladder, and its design reveals a beautiful story about how to build powerful theories by respecting the fundamental rules of the universe.

Let's begin our climb up "Jacob's Ladder," the hierarchy of density functional approximations.

At the bottom rung, we have the ​​Local Density Approximation (LDA)​​. You can think of this as looking at the electronic world with blurry vision. LDA approximates the energy at each point by assuming the electrons there behave like they are in a uniform "electron soup" of the same density. It's a simple, elegant idea, but it's too simple for the complex, non-uniform landscapes inside real atoms and molecules.

Taking a step up to the second rung brings us to the ​​Generalized Gradient Approximation (GGA)​​. This is like getting a new pair of glasses. A GGA not only sees the density $n(\mathbf{r})$ at a point but also how it's changing—its gradient, $\nabla n(\mathbf{r})$. This extra information allows GGAs to distinguish between regions of high and low density, and regions where the density is changing slowly or rapidly. It was a huge improvement, but GGAs have a critical blind spot.

Imagine two completely different chemical environments: the middle of a tight, localized covalent bond (like in a diamond crystal) and a region within a metal where electrons are delocalized and flow freely. It is entirely possible for these two distinct situations to have, just by chance, the very same density and the very same density gradient at a particular point. To a GGA functional, these two points are indistinguishable. It is blind to the underlying nature of the chemical bond. It's like looking at a gray patch and being unable to tell if it's a piece of solid rock or the fluffy fur of a cat. To see the difference, we need a more powerful lens.

This is where the third rung of the ladder comes in: the ​​meta-GGA​​. Meta-GGAs introduce a third, more discerning ingredient: the Kohn-Sham kinetic energy density, $\tau(\mathbf{r})$. Intuitively, $\tau(\mathbf{r})$ tells us not just how many electrons are at a point, but also how "agitated" or "wiggly" their underlying wavefunctions are. It's defined from the Kohn-Sham orbitals $\psi_i$ as:

Because it depends on the orbitals, $\tau(\mathbf{r})$ contains a wealth of information about the quantum mechanical structure of the system that is simply absent in $n(\mathbf{r})$ and $\nabla n(\mathbf{r})$ alone. It is our quantum zoom lens, capable of resolving the differences that a GGA misses.

Having a more powerful lens is one thing; knowing how to use it is another. The genius of the SCAN functional lies in how it deploys $\tau(\mathbf{r})$ to build a remarkably clever, point-by-point "bond-type detector." This detector is a dimensionless variable called the ​​iso-orbital indicator​​, $\alpha(\mathbf{r})$.

The definition of $\alpha$ is a masterpiece of physical intuition:

Let's dissect this expression piece by piece:

• $\tau(\mathbf{r})$ is the "true" kinetic energy density at point $\mathbf{r}$ for our system.

• $\tau_W(\mathbf{r}) = \frac{|\nabla n(\mathbf{r})|^2}{8n(\mathbf{r})}$ is the ​​von Weizsäcker kinetic energy density​​. This isn't just some arbitrary function. It has a profound physical meaning: it is the exact kinetic energy density a system would have if it were described by only a single spatial orbital. A beautiful, exact result of quantum mechanics shows that in any one-electron system (like a hydrogen atom), or any region of space dominated by a single orbital, the true kinetic energy density $\tau(\mathbf{r})$ becomes identically equal to $\tau_W(\mathbf{r})$.

• $\tau_{\mathrm{unif}}(\mathbf{r})$ is the kinetic energy density of a uniform electron gas (our "electron soup") that has the same density $n(\mathbf{r})$. This serves as our reference for a "many-orbital," highly delocalized system.

Now, look at what $\alpha$ is doing. It's a measure of how different the true kinetic energy is from the ultimate single-orbital limit, scaled by the many-orbital uniform gas limit. The value of $\alpha$ acts as a signal to the functional, telling it what kind of physics is dominant at each and every point in space:

• ​​$\alpha = 0$​​: This occurs when $\tau(\mathbf{r}) = \tau_W(\mathbf{r})$. The detector is screaming: "This is a single-orbital region!" This is the case for a hydrogen atom, a helium atom, or in the decaying tails of the electron density far from any molecule. It flags regions of extreme quantum localization.

• ​​$\alpha \approx 1$​​: This occurs when the density is very slowly varying. In this case, $\tau(\mathbf{r})$ approaches the uniform gas value $\tau_{\mathrm{unif}}(\mathbf{r})$, and $\tau_W(\mathbf{r})$ is very small. The detector signals: "This looks just like a uniform electron soup!" This is the characteristic signature of a simple metal.

• ​​$0 < \alpha < 1$​​: This is the vast middle ground, characteristic of the covalent bonds found in most molecules and semiconductors.

• ​​$\alpha \gg 1$​​: This surprising limit arises in regions of very low density where the wavefunctions of multiple fragments weakly overlap, for instance, between two noble gas atoms that are "touching." The detector is saying: "This is a tenuous, multi-orbital region of weak interaction. Handle with care."

In essence, $\alpha$ acts as a local "bond character" analyzer. Armed with this extraordinary tool, the SCAN functional can finally overcome the GGA's blindness, distinguishing a covalent bond from a metallic one even if their $n$ and $\nabla n$ are identical.

Now that we have this amazing detector, how do we build the final recipe? Many functionals are developed by fitting their mathematical form to a large database of experimental chemical properties—they are "empirically" tuned. The SCAN philosophy is radically different. Its name says it all: ​​Strongly Constrained and Appropriately Normed​​. Rather than being fitted to data, SCAN is constructed from first principles to satisfy a list of 17 known ​​exact constraints​​—rigorous mathematical conditions that the true, universal functional is known to obey.

The $\alpha$ indicator is the key that allows SCAN to satisfy different constraints in different physical regimes:

• ​​The One-Electron Constraint ($\alpha=0$)​​: The exact functional must be free of "self-interaction"—an electron should not interact with itself. For any one-electron system, the exchange energy must exactly cancel the spurious self-Hartree energy. SCAN is built to enforce this condition perfectly whenever it detects an $\alpha=0$ region. This is why SCAN is exact for the hydrogen atom, a feat most simpler functionals cannot achieve.

• ​​The Slowly Varying Limit ($\alpha \approx 1$)​​: For a density that varies very slowly in space, the exact functional's behavior is known from many-body theory as a "gradient expansion." SCAN is constructed to reproduce this expansion correctly in the $\alpha \approx 1$ regime. This is one of its "appropriate norms."

• ​​The Jellium Surface Norm​​: SCAN is also "normed" to describe a difficult test case correctly: the surface of an idealized metal, or "jellium." This is a system with a very large and rapid change in electron density. By being designed to get the surface energy of this system right, SCAN becomes more accurate for real-world surfaces, interfaces, and adsorption.

• ​​The Weak Interaction Regime ($\alpha \gg 1$)​​: Here lies one of SCAN's most remarkable achievements. The weak, "sticky" forces between molecules, known as ​​van der Waals​​ or dispersion forces, are fundamentally a nonlocal correlation effect. A semilocal functional like SCAN shouldn't be able to describe them. Yet, by carefully designing the functional to obey all known constraints, an amazing thing happens. In the $\alpha \gg 1$ regions corresponding to weak orbital overlap, SCAN's constrained form produces an attractive interaction that effectively mimics intermediate-range van der Waals forces. It's an emergent property—a reward for diligently following the rules of the universe.

By using $\alpha$ to interpolate seamlessly between these different constrained behaviors, SCAN becomes a "functional for all seasons." It can accurately describe the tight covalent bonds in a diamond, the delocalized metallic bonding in a simple metal, and the weak non-covalent interactions that hold biological molecules together. This versatility is showcased in its ability to solve long-standing problems, such as correctly predicting the pressure-induced phase transition of silicon from a covalent semiconductor to a metal—a task where simpler GGAs systematically fail because they cannot distinguish the two bonding types as effectively.

This incredible power and complexity do not come for free. The very same mechanism that gives SCAN its discerning power—the $\alpha$ indicator and the switching functions that interpolate between different physical regimes—makes the functional's energy landscape mathematically "bumpy" and complex.

For a computer performing a DFT calculation, this means that accurately calculating the total energy with SCAN requires extreme numerical precision. It's like trying to measure the area of a craggy, mountainous terrain. A coarse measuring grid (the computational equivalent of using a large, clumsy yardstick) will miss all the sharp peaks and valleys, leading to a wrong answer. To get an accurate result with SCAN, one must use exceptionally dense numerical integration grids, a fine-toothed comb that can capture every intricate feature of the energy landscape. This numerical sensitivity has even spurred the development of "regularized" versions, like ​​r2SCAN​​, which gently smooth out the sharpest bumps in the functional to make calculations more stable, while meticulously preserving all of the original 17 exact constraints.

The journey to SCAN shows us a path forward in the quest for the universal functional. It is not the final answer—no semilocal functional can ever be truly exact, as some quantum phenomena are fundamentally nonlocal. But it is a monumental achievement, a testament to the power of building theories not from empirical fitting, but from deep physical intuition and a profound respect for the fundamental constraints that govern our quantum world. It teaches us that by asking the right questions and identifying the right ingredients, we can create theories of surprising power and universality.

We have spent some time getting to know the inner workings of the SCAN functional, appreciating the elegant and rigorous physical constraints that serve as its blueprint. We saw how it was built not by fitting to data, but by listening to the exact rules that nature herself must obey. Now, the real fun begins. Like being handed the keys to a new kind of vehicle, the natural question is: where can it take us? In this chapter, we will embark on a journey across the vast landscapes of modern science to see what worlds the SCAN functional unlocks, from the intricate dance of atoms in a catalyst to the deep mysteries of exotic materials.

For a chemist, the world is made of molecules. To understand this world, we must first be able to predict its most fundamental properties: how atoms arrange themselves into stable structures, and how those structures transform during a chemical reaction.

Getting the geometry right is the essential first step. For simple organic molecules, many theoretical tools do a reasonably good job. But chemistry is full of delightful troublemakers, and the transition metals are chief among them. Their complex d-electrons make for a rich and varied chemistry, but they are a notorious headache for theoretical models. For decades, chemists have relied on hybrid functionals like B3LYP, which mix in a bit of "exact" exchange and are parameterized to get good results for many common systems. They work, often quite well, but one might feel a bit like a chef using a pre-made spice blend—it’s effective, but you don't fully understand why each ingredient is there.

SCAN offers a different philosophy. By building from the ground up with physical constraints, it aims to be a more universal tool. When we point it at challenging transition metal complexes, we often find that it predicts metal-ligand bond lengths with remarkable accuracy, frequently outperforming older, more heavily parameterized methods. This is not a magic trick; it is the direct payoff of a functional that is better at describing the diverse types of chemical bonds that these metals can form.

Once we are confident in our structures, we can ask about reactions. The heart of a chemical reaction is the transition state—a fleeting, high-energy arrangement of atoms that sits at the peak of the energy barrier separating reactants from products. The height of this barrier determines the reaction rate. Accurately predicting it is one of the holy grails of computational chemistry. This is where SCAN's clever design truly shines.

Recall the iso-orbital indicator, $\alpha(\mathbf{r})$, the functional's built-in "bond-type detector." We saw that in regions dominated by a single orbital, such as the stretched bond in a transition state or the tail of an atom, $\alpha(\mathbf{r})$ is exactly zero. SCAN is designed to recognize this and, in that limit, to become exact for one-electron systems, a property that drastically reduces the pernicious "self-interaction error" that plagues many other functionals. This error is like a ghost in the machine, an artifact where an electron incorrectly interacts with itself, and it is particularly damaging when describing stretched bonds. By satisfying this one-electron constraint, SCAN provides a more physically sound description of reaction barriers compared to many of its predecessors, leading to more reliable predictions of chemical reactivity.

Leaving the chemist's flask, we now enter the vast and orderly world of crystalline solids. Here, atoms are not in small, isolated groups, but are arranged in endless, repeating lattices. SCAN's talent for describing different bonding environments makes it a powerful tool for the materials physicist as well.

Let's start with the most basic questions: How far apart are the atoms in a crystal, and how stiff is the material? SCAN provides excellent predictions for the lattice constants—the fundamental repeating distances in a crystal—for a wide range of materials, from simple semiconductors like silicon to metals like iron. This is a direct consequence of its balanced description of exchange and correlation effects across different density regimes.

But a crystal is not a static object. Its atoms are constantly quivering in a collective, quantized dance. These vibrations are called phonons, and they determine properties like a material's heat capacity and thermal conductivity. Using SCAN, we can compute the frequencies of these phonons with high accuracy. The calculation essentially links a macroscopic, measurable property—like the bulk modulus, which tells us how a material resists compression—to the microscopic springiness of the chemical bonds holding the atoms together.

Now, let's turn up the heat with an even more challenging phenomenon: magnetism. In materials like iron, the quantum mechanical spin of the electrons align, creating a collective magnetic moment. This is a subtle correlation effect that is notoriously difficult to capture. Simpler functionals like LDA and PBE often struggle, either failing to predict magnetism where it exists or getting the size of the magnetic moment wrong. SCAN, by providing a more refined description of electron exchange, often gives a significantly better account of the magnetic properties of metals, bringing theory one step closer to the reality of these fascinating materials.

So far, we have focused on the strong chemical bonds that hold molecules and solids together. But there is another world of interactions, a subtler and more ephemeral one, governed by the weak forces collectively known as van der Waals (vdW) interactions. These are the forces that hold layers of graphite together, allow geckos to walk on ceilings, and shape the structure of DNA.

These long-range forces arise from the correlated fluctuations of electron clouds. Imagine two neutral, spherical atoms far apart. On average, they have no dipole moment. But at any instant, the electron cloud on one atom might fluctuate, creating a temporary dipole. This fleeting dipole creates an electric field that induces a corresponding dipole in the second atom, and the two temporary dipoles attract each other. This is the London dispersion force.

Here, we encounter a fundamental limitation. SCAN, for all its sophistication, is a semilocal functional. This means the energy at a point $\mathbf{r}$ depends only on the electronic properties at or very near that same point. The vdW interaction, however, is fundamentally nonlocal—it's a correlation between what's happening at one point and another far away. Consequently, no semilocal functional, SCAN included, can capture the correct long-range $1/R^6$ decay of the dispersion force. For two weakly interacting atoms, SCAN predicts an interaction that dies off exponentially, which is physically incorrect.

Does this mean SCAN is useless for these systems? Not at all. It turns out that at the intermediate distances where vdW interactions are often most important, SCAN's advanced design allows it to capture some of the attractive "dispersion-like" energy, far more than simpler GGAs. Its ability to recognize the weak-overlap region between two atoms (where the indicator $\alpha$ becomes very large) enables it to provide a surprisingly reasonable description of binding in some vdW systems.

For true accuracy, however, we must explicitly add the missing physics. This is the idea behind methods like SCAN+rVV10 or SCAN+D3. A truly nonlocal term, designed to reproduce the correct long-range physics, is added to the SCAN functional. The result is a beautiful synergy: SCAN handles the short- and medium-range interactions with its sophisticated semilocal machinery, while the nonlocal correction seamlessly takes over at long range. This combined approach is essential for accurately describing layered materials like graphite and MoS$_2$, which are the foundation of 2D materials science. Predicting their exfoliation energy—the energy required to peel one layer off—demands a theory that gets both the short-range repulsion and the long-range vdW "glue" just right.

Armed with this powerful and versatile functional, we can now venture to the frontiers of modern research. One such frontier is heterogeneous catalysis, where chemical reactions are sped up on the surfaces of materials. Designing better catalysts for clean energy or industrial processes is a major goal of modern science. Here, DFT calculations, and SCAN in particular, are indispensable tools. They allow us to study how molecules adsorb onto a catalyst surface, a critical step in any catalytic cycle. SCAN's balanced description of different bond types, combined with dispersion corrections, provides a more reliable picture of adsorption energies and surface reactions than ever before.

Perhaps the ultimate challenge for electronic structure theory lies in a class of materials known as strongly correlated systems. These are materials where the simple picture of independent electrons breaks down completely. Mott insulators, like nickel oxide (NiO), are the canonical example. Based on simple band theory, they "should" be metals, but strong on-site repulsion between electrons forces them into an insulating state. These materials are at the heart of phenomena like high-temperature superconductivity.

Predicting the properties of Mott insulators is extraordinarily difficult. Simple functionals like LDA and PBE famously fail, incorrectly predicting them to be metals. SCAN, by being less prone to the self-interaction error that plagues simpler functionals, provides a much better starting point. While it may not always be enough on its own to open the full energy gap, it provides a more physical description of the electronic structure, making it a superior foundation upon which more advanced (and computationally expensive) theories like DFT+U or hybrid functionals can be built.

Our journey has taken us from simple molecules to the most complex and mysterious materials known. At every step, we've seen how the SCAN functional, by staying true to the fundamental constraints of quantum mechanics, provides a more robust, reliable, and universal lens for viewing the electronic world. It is a testament to the power of physics-based design and a significant milestone in the ongoing quest to create a truly universal map of the quantum landscape.