Why Projection-Based DMRG-in-DFT Cannot Be Exact, Even with the Exact Exchange-Correlation Functional

This paper establishes that projection-based DMRG-in-DFT embedding is inherently nonvariational and fundamentally limited by nonadditive exchange-correlation errors at the subsystem-environment interface, which cannot be resolved even by employing exact functionals or pair-density approaches.

The Gist

Here is an explanation of the paper, translated into everyday language with some creative analogies.

The Big Picture: The "Too Big to Solve" Problem

Imagine you are trying to solve a massive, incredibly complex jigsaw puzzle. This puzzle represents a molecule. Some parts of the puzzle are simple and easy to fit together (like the edges), but the center is a chaotic mess of pieces that all look the same and snap together in weird ways. This "messy center" is what scientists call strong electron correlation.

• The Old Way (Wavefunctions): To solve the messy center perfectly, you need a super-precise method (like DMRG). But this method is so computationally heavy that if you try to use it on the whole puzzle, your computer explodes. You can only solve tiny puzzles this way.
• The Other Way (DFT): There is a faster, cheaper method (Density Functional Theory, or DFT) that can solve huge puzzles. But it's like using a blurry camera; it works great for simple pictures but gets the messy center completely wrong.

The Goal: Scientists want to use the "super-precise" method just for the messy center and the "fast, blurry" method for the rest. This is called Embedding. You treat the center with high precision and the surroundings with the fast method, hoping they work together perfectly.

The Paper's Discovery: The "Perfect" Plan Has a Flaw

The authors of this paper looked at a specific, popular embedding technique called Projection-Based DMRG-in-DFT. The idea was: "Let's force the precise center and the fast surroundings to be mathematically orthogonal (like two people standing in different rooms who can't touch)."

They asked a fundamental question: "If we had a magic, perfect formula for the 'blurry' method, would this embedding technique give us the exact, perfect answer?"

The Answer: No.

Even with a perfect formula, this specific method is mathematically broken. It is "non-variational," which is a fancy way of saying it doesn't play by the rules of energy conservation. It's like a scale that is broken in a specific way: it will always tell you that an object weighs less than it actually does.

The Analogy:
Imagine you are building a house. You hire a master architect for the foundation (the active region) and a handyman for the roof (the environment).

• The Exact Theory says: "To get the perfect house, you must account for the exact amount of friction between the foundation and the roof."
• The Projection-Based Method says: "Let's just pretend the foundation and roof don't touch at all to make the math easier."
• The Result: By pretending they don't touch, you accidentally remove a necessary "friction cost" from your budget. Your final calculation says the house is cheaper (lower energy) than it really is. No matter how good your architect or handyman is, the math is fundamentally flawed because you ignored that friction.

The Real Villain: The "Ghost" Interaction

The authors didn't just stop at finding the math flaw. They asked, "Okay, the math is slightly off, but is that the main reason our results are wrong when we stretch chemical bonds?"

They tested this on two scenarios:

1. A chain of Hydrogen atoms being pulled apart.
2. A Propionitrile molecule with a bond being stretched.

The Surprise:
They expected the main error to come from the "fractional spin error" (a common bug in DFT where it gets confused about electron spins). They thought, "If we fix the spin error, we should get perfect results!"

They tried using a new, fancy tool called PDFT (Pair-Density Functional Theory) which is designed to have zero spin error.

The Result: It got worse.

Why?
The real problem isn't the spin; it's the interface.
When you stretch a bond, the electrons from the "active" part (the messy center) start to leak out and overlap with the "environment" (the surroundings).

• The current methods (like PBE and even PDFT) are terrible at calculating the energy of this overlap.
• They underestimate how much the two parts "hate" each other (or rather, how they interact quantum mechanically).
• Because they underestimate this interaction, they think the broken bond is more stable than it actually is.

The Analogy:
Imagine two magnets (the active part and the environment).

• The Spin Error is like a magnet being slightly the wrong color.
• The Interface Error is like the magnets being glued together, but your glue calculator says they aren't sticking at all.
• Even if you paint the magnets the perfect color (fixing the spin error), if your glue calculator is broken, you will still think the magnets will fly apart when they actually stick together. The "glue" (the non-additive exchange-correlation energy) is the real problem.

The Conclusion

1. The Method is Flawed: The specific "Projection-Based" method used today is mathematically incapable of being 100% exact, even in a perfect world, because it ignores a subtle kinetic energy cost.
2. The Spin Error is a Red Herring: Fixing the "spin" problems in the formulas (using PDFT) doesn't help because that wasn't the main issue.
3. The Real Fix: The biggest source of error is how the active region and the environment talk to each other when their electron clouds overlap. We need better formulas to calculate this specific "handshake" energy.

In short: We are trying to solve a puzzle by using a super-precise tool for the middle and a cheap tool for the edges. The paper tells us that the way we are currently connecting these two tools is broken, and simply making the cheap tool "smarter" about spins won't fix it. We need to fix the connection point itself.

Technical Summary

Here is a detailed technical summary of the paper "Why Projection-Based DMRG-in-DFT Cannot Be Exact, Even with the Exact Exchange-Correlation Functional" by Monino et al.

1. Problem Statement

The paper addresses the theoretical limitations of Projection-Based Wavefunction-in-DFT (PB-WF-in-DFT) embedding, specifically when using the Density Matrix Renormalization Group (DMRG) for the active subsystem.

• Context: Embedding methods aim to combine high-level wavefunction (WF) methods (like DMRG) for strongly correlated regions with Density Functional Theory (DFT) for the environment to treat large systems.
• The Gap: While Projection-Based DFT (PB-DFT) is known to be exact when both subsystems are treated with DFT, it was unclear whether the hybrid WF-in-DFT formulation (where the active region is a correlated WF and the environment is KS-DFT) is formally exact.
• The Question: Can PB-WF-in-DFT recover the exact ground-state energy if the exact exchange-correlation (xc) functional were used?
• Practical Issue: In practice, these methods often yield errors in dissociating bonds. It is unclear if these errors stem from the projection approximation itself or from the approximate nature of standard xc functionals (e.g., fractional-spin errors).

2. Methodology and Theoretical Framework

A. Theoretical Derivation

The authors construct a rigorous theoretical framework to compare the exact embedding functional with the approximate projection-based functional.

1. Exact Functional Construction: They define an exact energy functional, $E_{\text{WF-in-DFT}}$, for a correlated wavefunction $\Psi_{NA}$ embedded in a Kohn-Sham (KS) orbital environment.
   • This functional includes a term representing the difference between the kinetic energy of the active orbitals constrained to be orthogonal to the environment and the non-interacting kinetic energy functional of the active density ($\Delta T_s$).
   • Key Finding: The exact functional requires knowledge of this unknown kinetic energy difference term:
     $$\Delta T_s[\rho_A, \{\phi^{KS}_i\}_{NB}] = \min_{\{\phi_i\} \perp \{\phi^{KS}_i\}} T[\{\phi_i\}] - T_s[\rho_A] \geq 0$$
2. Comparison with Projection-Based Formulation: The standard projection-based functional (Miller et al.) is shown to be equivalent to the exact functional minus the $\Delta T_s$ term.
   • Consequence: Because $\Delta T_s \geq 0$, the standard projection-based functional is non-variational. Its minimum is bounded from above by the exact ground-state energy ($E_0$), meaning it systematically underestimates the energy (yields $E < E_0$) when the active and environment densities overlap.

B. Computational Approach

To analyze the sources of error in practical applications, the authors performed benchmark calculations on two systems:

1. Dissociating $H_{20}$ Chain: A linear chain of 20 hydrogen atoms with a central active fragment (4 or 8 atoms) undergoing bond stretching.
2. Propionitrile ($CH_3CH_2CN$): The dissociation of the C$\equiv$N triple bond.

• Methods:
  • Active Region: DMRG (using MOLMPS) with the cc-pVDZ basis set.
  • Environment: KS-DFT (using ORCA).
  • Functionals: Standard PBE (semilocal) and Pair-Density Functional Theory (PDFT).
• Strategy: They compared DMRG-in-DFT results against high-level DMRG benchmarks to isolate errors in the nonadditive exchange-correlation ($E^{nadd}_{xc}$) and kinetic energy terms. They specifically tested if replacing PBE with PDFT (which eliminates fractional-spin errors) would fix the embedding errors.

3. Key Contributions

1. Proof of Non-Exactness: The paper provides the first formal proof that standard projection-based WF-in-DFT embedding is inherently non-exact, even with the exact xc functional. The omission of the non-negative kinetic energy difference term ($\Delta T_s$) makes the method non-variational.
2. Identification of Dominant Error Source: Through analysis of dissociating bonds, the authors demonstrate that the error introduced by neglecting $\Delta T_s$ is negligible compared to the error arising from the nonadditive exchange-correlation energy ($E^{nadd}_{xc}$).
3. Mechanism of Failure: The primary error stems from the self-interaction error (SIE) and the delocalization error at the subsystem-environment interface.
   • In dissociating systems, the active fragment's density delocalizes into the environment.
   • Approximate functionals (PBE) fail to correctly describe the coupling between the active and environment densities, leading to an insufficiently negative nonadditive xc energy.
4. Inefficacy of PDFT: The study shows that using PDFT (which is free of fractional-spin errors) does not remedy the embedding errors. In fact, PDFT often performs slightly worse than PBE in this context because the error is driven by SIE at the interface, not fractional-spin errors.

4. Results

• Energy Behavior: In the dissociation limit of the $H_{20}$ chain, DMRG-in-PBE yields energies above the benchmark (positive error), whereas the theoretical non-variationality suggests it should be below (negative error). This confirms that the xc error (positive) overwhelms the kinetic energy error (negative).
• Error Decomposition:
  • For the $H_{20}$ chain, the error in the nonadditive xc energy is dominated by the difference between the fractional-spin error (FSE) of the composite system and the self-interaction error (SIE) of the embedded fragment.
  • The errors do not cancel out because the nature of the error in the active fragment (SIE due to delocalization) differs from the error in the composite system (FSE).
• PDFT Performance: Replacing PBE with PDFT for the nonadditive term did not reduce the error. The PDFT functional also suffers from SIE in the delocalized regime, and since the error cancellation mechanism fails, the total error remains large (approx. 56 mHa for $H_{20}$ vs. 34 mHa for PBE).
• Mitigation Strategies:
  • Enlarging the Active Region: Increasing the size of the active fragment (e.g., from 4 to 8 H atoms) reduced the error by suppressing density overlap variations.
  • Increasing Separation: Increasing the distance between the active fragment and the environment reduced the error by ~25 mHa, confirming that the error is driven by the overlap/coupling between subsystems.

5. Significance and Conclusion

• Theoretical Impact: The work clarifies that projection-based embedding is not a "black box" exact theory. It relies on an approximation (neglecting $\Delta T_s$) that introduces a systematic bias, though this bias is often overshadowed by functional inaccuracies in practical scenarios.
• Practical Guidance:
  • Simply using "better" functionals that fix fractional-spin errors (like PDFT) is not sufficient to fix DMRG-in-DFT embedding for strongly correlated, dissociating systems.
  • The bottleneck for accuracy lies in the nonadditive exchange-correlation functional.
• Future Directions: The authors suggest that developing specific corrections for the nonadditive xc energy (e.g., via noninteracting xc corrections or adiabatic connection methods) is essential for achieving high accuracy in WF-in-DFT embedding. The results indicate that with a properly treated nonadditive xc functional, DMRG-in-DFT can attain high accuracy despite the intrinsic limitations of the projection scheme.

In summary, the paper establishes that the failure of projection-based DMRG-in-DFT in dissociating systems is not primarily due to the projection approximation itself, but rather due to the inability of standard (and even advanced) xc functionals to correctly describe the non-classical coupling and self-interaction effects at the interface between the active wavefunction and the DFT environment.