Shannon and Rényi entropies of molecular densities: insights into extensivity and the incomplete description of electron correlation

This paper demonstrates that Shannon and Rényi entropies derived from electron densities and shape functions are unreliable descriptors of static electron correlation and extensivity, as they fail to accurately encode correlation effects and often violate extensivity, suggesting that robust entropic measures must instead be constructed from higher-dimensional Hilbert-space objects.

The Gist

The Big Picture: Trying to Measure a Crowd's "Chaos"

Imagine you are a scientist trying to understand how a group of people (electrons) behaves in a room (a molecule). You want to measure how "spread out" or "chaotic" they are. In the world of information theory, this measurement is called Entropy.

For a long time, chemists have tried to use two specific tools to measure this chaos:

1. Shannon Entropy: Like measuring the total noise level in a room.
2. Rényi Entropy: A more complex version that weighs loud noises differently than quiet whispers.

The big hope was that these tools could tell us exactly how well the electrons are "getting along" (a concept called electron correlation). If the electrons are perfectly coordinated, the entropy should be low. If they are chaotic and independent, it should be high.

The Paper's Conclusion: The authors of this paper ran a series of tests and found a major problem. These tools are actually bad at their job. They fail to tell the difference between a well-organized crowd and a chaotic one, and they break the fundamental rules of physics when you try to measure large groups.

---

The Analogy: The "Two-Atom" Dance

To test their tools, the authors looked at the simplest possible dance: two atoms moving apart (dissociating). Imagine two dancers holding hands. As they walk away from each other, eventually they let go and become two separate, independent people.

In a perfect world, if you measure the "chaos" of the pair while they are holding hands, and then measure the chaos of the two individuals separately, the math should add up perfectly.

• Rule of Extensivity: The chaos of the whole group should equal the sum of the chaos of the individuals. (If you have two identical rooms, the total mess should be exactly double the mess of one room).

The Three Problems They Found

The authors discovered three major flaws in using these entropy tools for chemistry:

1. The "Blindfolded" Problem (Missing Static Correlation)

The Issue: When two atoms are far apart, they often need to share a "secret handshake" (static correlation) to stay stable.
The Analogy: Imagine two dancers who, when far apart, must pretend to be holding hands to avoid falling over.

• The Flaw: The Shannon entropy tool is like a blindfolded observer. It looks at the crowd and sees the same amount of "noise" whether the dancers are holding hands perfectly or just standing there awkwardly. It cannot detect the subtle "static correlation" that keeps the molecule stable.
• Result: It fails to tell the difference between a good description of the molecule and a bad one.

2. The "Broken Scale" Problem (Violation of Extensivity)

The Issue: When the atoms get very far apart, the math stops adding up.
The Analogy: Imagine you have a scale that weighs a single apple as 100 grams. If you put two apples on the scale, it should read 200 grams. But in this paper, the "Shape Function" entropy tool is like a broken scale that reads 200 grams for one apple, but 250 grams for two apples.

• The Flaw: The tool adds an extra, phantom "weight" (a mathematical term called a logarithmic term) just because there are two atoms. It doesn't matter how far apart they are; the tool insists the total chaos is more than the sum of the parts.
• Result: This violates a fundamental rule of physics called extensivity. You can't use a ruler that changes its own length depending on how many objects you measure!

3. The "Over-Confident" Problem (Hartree-Fock vs. Reality)

The Issue: The authors compared "simple" calculations (Hartree-Fock) with "perfect" calculations (Full Correlation).

• The Flaw: The simple calculation (Hartree-Fock) is like a student who guesses the answer without doing the hard math. Surprisingly, this "guessing" method produced higher entropy numbers than the "perfect" method.
• Result: Usually, we expect the "perfect" method to show more complexity. Here, the imperfect method looked more chaotic than the perfect one. This suggests the tool is measuring the wrong things (like the shape of the basis set) rather than the actual electron behavior.

The "Shape Function" Trap

The authors also tested a variation called the Shape Function. This is like taking the electron density and normalizing it (making the total number of electrons equal to 1, like a probability map).

• The Metaphor: Imagine you have a map of a city. The Shannon entropy measures the total population density. The Shape Function measures the pattern of the city, ignoring the population size.
• The Problem: While this sounds nice, the authors found that when you use the Shape Function, the "broken scale" problem gets even worse. It introduces a permanent error that makes it impossible to compare molecules of different sizes fairly.

The Final Verdict: What Should We Do?

The paper concludes that electron density is not enough.

Think of electron density as a 2D shadow of a 3D object. You can see the outline, but you miss the depth. The authors argue that to truly understand how electrons correlate (how they interact and dance together), we need to look at the full 3D object (the wavefunction or higher-dimensional Hilbert space objects).

In simple terms:
Trying to understand the complex behavior of electrons just by looking at their "shadow" (density) using these entropy tools is like trying to understand a symphony by only looking at the sheet music's volume markings. You miss the harmony, the timing, and the true interaction between the instruments. We need better tools that look at the whole orchestra, not just the shadow.

Technical Summary

1. Problem Statement

The paper addresses the reliability of information-theoretic measures, specifically Shannon ($S$) and Rényi ($S_\alpha$) entropies derived from the electron density ($\rho$) and the shape function ($\sigma = \rho/N$), as descriptors of electronic correlation and chemical bonding.

While these entropic measures are widely used in theoretical chemistry to analyze molecular structure, reactivity, and complexity, the authors question their ability to:

1. Capture Static Correlation: Do these measures correctly reflect the degree of static correlation (crucial for bond dissociation) present in the underlying wavefunction?
2. Satisfy Extensivity: Do these measures behave additively (extensively) when a molecule dissociates into non-interacting fragments at infinite internuclear distance?

The authors hypothesize that density-based entropies may fail to encode static correlation and may violate extensivity, particularly when using the shape function or generalized Rényi entropies ($\alpha \neq 1$).

2. Methodology

The study employs a combination of rigorous algebraic derivation and numerical integration across various molecular systems and theoretical levels.

• Theoretical Framework:

  • Partitioning Scheme: The authors utilize a Mulliken-like atomic partition of the electron density. The total density $\rho(r)$ is decomposed into atomic ($\rho_{AA}$) and interatomic/overlap ($\rho_{AB}$) contributions based on a primitive basis set expansion.
  • Entropy Decomposition: They derive a rigorous decomposition of the Shannon entropy density into additive (net atomic and overlap) and non-additive contributions. This involves rewriting the logarithm of the sum of densities to isolate coupling terms.
  • Generalization to Rényi: The partitioning is extended to Rényi entropies of order $\alpha$, which involve integrals of $\rho(r)^\alpha$.

• Numerical Models:

  • Systems: Two-electron homonuclear diatomics ($H_2$) and larger systems ($N_2$, $H_2O$).
  • Levels of Theory:
    • Hartree-Fock (HF): Single-determinant, lacks static correlation (incorrectly dissociates).
    • Heitler-London (HL): Minimal basis, includes static correlation (correctly dissociates).
    • Full Configuration Interaction (FCI) / Complete Active Space (CAS): Multideterminant methods providing correct static correlation.
  • Basis Sets: Minimal basis (STO-6G) and extended basis (aug-cc-pVTZ).
  • Limits: Analysis focuses on the infinite internuclear distance limit ($R \to \infty$) to assess dissociation behavior and extensivity.

• Computational Tools:

  • Geometry optimizations and energies calculated using Gaussian16 and GAMESS.
  • Numerical integration of entropy terms performed using an in-house code (PROMOLDEN) employing Becke's multicenter integration scheme.

3. Key Contributions

1. Rigorous Algebraic Partitioning: The authors provide a formal decomposition of Shannon and Rényi entropies into additive (atomic) and non-additive (coupling) terms, allowing for a precise analysis of how these quantities behave as atoms separate.
2. Identification of Extensivity Violations: They demonstrate that while the electron-density Shannon entropy ($S_\rho$) satisfies extensivity in minimal bases, the shape-function entropy ($S_\sigma$) and Rényi entropies ($\alpha \neq 1$) inherently violate extensivity due to persistent non-additive terms in the dissociation limit.
3. Decoupling of Entropy from Static Correlation: The study proves that for minimal basis sets, the Shannon entropy converges to the same limit regardless of whether the wavefunction includes static correlation (HF vs. HL/FCI), rendering it blind to the quality of the wavefunction's dissociation.
4. Basis Set Dependence: The authors show that in extended basis sets, uncorrelated HF densities yield significantly higher entropies than correlated densities at large distances, a result driven by variational deformation of atomic densities rather than physical correlation effects.

4. Key Results

A. Minimal Basis Sets (STO-6G)

• Shannon Entropy ($S_\rho$): At $R \to \infty$, $S_\rho$ converges to the sum of the isolated atomic entropies ($2S_{atom}$) for all methods (HF, HL, FCI).
  • Implication: $S_\rho$ fails to distinguish between a correctly dissociating wavefunction (HL/FCI) and an incorrectly dissociating one (HF). It does not encode the amount of static correlation.
• Shape Function Entropy ($S_\sigma$): At $R \to \infty$, $S_\sigma$ converges to $S_{atom} + \log(N)$ (where $N$ is the number of electrons).
  • Implication: This violates extensivity. The entropy of the dissociated system is not the sum of the entropies of the parts; it contains a spurious $\log(N)$ term.
• Rényi Entropy ($S_\alpha, \alpha \neq 1$): Even for the electron density, these entropies fail to reach the simple atomic sum. They retain a persistent non-additive contribution dependent on $\alpha$ and the number of centers, preventing them from describing the correct atomic limit.

B. Extended Basis Sets (aug-cc-pVTZ)

• HF vs. Correlated Methods: In extended bases, the HF approximation introduces spurious ionic character at large distances, leading to "deformed" atomic densities. Consequently, the HF entropy saturates at a value significantly higher than the correct atomic limit.
• Correlated Methods (CAS/FCI): Only sufficiently correlated methods (e.g., full-valence CAS) recover the correct atomic limit (or the correct deformed limit consistent with the method).
• Conclusion: In extended bases, $S_\rho$ can distinguish between methods, but not because it measures correlation directly; rather, it measures the variational deformation of the density. Uncorrelated methods overestimate entropy due to this deformation.

C. Non-Additive Contributions

• The non-additive entropy term ($S_{nadd}$) acts as a correction to the additive part.
• In the dissociation limit, $S_{nadd}$ vanishes for Shannon entropy in minimal bases but persists for Rényi entropies and shape functions.
• The authors suggest that the behavior of $S_{nadd}$ near equilibrium might be a better indicator of bonding than the total entropy, but it fails to describe the asymptotic limit correctly.

5. Significance and Conclusions

• Limitations of Density-Based Descriptors: The paper concludes that electron-density-based entropic measures are insufficient for capturing static correlation or providing a robust, extensive description of molecular dissociation. They fail to distinguish between correct and incorrect wavefunctions in minimal bases and exhibit non-extensive behavior in shape functions and Rényi entropies.
• Extensivity Challenge: The violation of extensivity in shape-function and Rényi entropies poses a fundamental barrier to using them as universal descriptors for systems of varying sizes.
• Future Directions: The authors argue that robust entropic descriptors for electronic correlation must be constructed from higher-dimensional Hilbert-space objects (e.g., the 1-particle reduced density matrix or the full wavefunction) rather than the 3D electron density alone. This would allow for the proper inclusion of phase information, entanglement, and momentum-space spread, which are lost in the density-only formulation.

In summary, the work serves as a critical re-evaluation of information-theoretic tools in quantum chemistry, demonstrating that while useful for certain structural trends, they are fundamentally flawed for describing the nuances of electron correlation and dissociation limits when based solely on the electron density.