Post-pulse dipole instability in adiabatic TDDFT: fact or artifact?

This paper demonstrates that the reported post-pulse dipole instability in adiabatic real-time TDDFT is a numerical artifact caused by incorrect non-linearities in the propagation scheme, which are absent when the same approximation is applied within the response-reformulated RR-TDDFT framework.

The Gist

The Big Question: Is the Computer Glitching?

Imagine you are watching a simulation of a molecule (specifically, a nitrogen molecule, $N_2$) getting hit by a super-fast, high-energy flash of light (an XUV pulse).

In recent computer simulations, scientists noticed something weird happening after the light flash was turned off. The molecule's "dipole" (a measure of how its electric charge is wiggling) was supposed to settle down and go quiet. Instead, after a few seconds of silence, it suddenly started wiggling violently again, growing stronger and stronger in a wild, exponential burst.

The scientists who found this called it a "dipole instability." They wondered: Is this a real physical phenomenon that happens in nature, or is it just a glitch in the computer code?

This paper says: It's a glitch. It is an "artifact" created by the way the computer was solving the math, not something that actually happens in the real world.

The Two Ways of Doing the Math

To figure this out, the authors ran the same simulation using two different mathematical "recipes" (formulations) for Time-Dependent Density Functional Theory (TDDFT). Think of these as two different ways to navigate a maze.

1. The Traditional Recipe (TDKS): This is the standard, most common way scientists have been doing this for years. It's like trying to drive a car by looking only at the road directly in front of your bumper right now, ignoring where you've been or where you're going. It makes a lot of assumptions to keep things simple.
2. The New Recipe (RR-TDDFT): This is a newer, more rigorous method. It's like having a GPS that remembers your entire route and calculates your path based on a complete map of the terrain, rather than just the spot under your tires.

The Experiment: The "Echo" That Shouldn't Exist

The researchers set up a race between these two recipes using the nitrogen molecule and the same XUV light flash.

• The Traditional Recipe (TDKS): Just like in the previous studies, this method showed the "dipole instability." After the light stopped, the molecule went quiet, then suddenly started screaming (oscillating wildly) on its own.
• The New Recipe (RR-TDDFT): When they used the new, more accurate recipe with the exact same settings, the instability disappeared completely. The molecule wiggled a bit while the light was on, and then settled down quietly afterward, exactly as physics would predict.

The Conclusion: Since the new, more accurate method didn't show the instability, the wild wiggling seen in the old method must be a fake side-effect of the math, not real physics.

Why Did the Old Method Fail? (The "Self-Driving" Analogy)

The paper explains why the old method failed using a concept called "memory."

• The Problem: The traditional method uses an "adiabatic approximation." In plain English, this means the computer calculates the forces on the electrons based only on the electron's position at this exact split-second. It has no memory of the past.
• The Glitch: Imagine you are pushing a child on a swing. If you push exactly when the swing is at the bottom, you add energy. If you push when it's at the top, you stop it.
  • In the real world (and the new math), the forces adjust smoothly.
  • In the old math, because it only looks at the "now," it accidentally pushes the swing at the perfect moment to make it go higher every single time. It creates a feedback loop where the system "drives itself."
  • The computer sees a tiny, natural wobble, and because of its "no memory" rule, it accidentally amplifies that wobble into a massive, impossible explosion of energy.

The Role of the "Absorbing Boundary"

The paper also highlights a crucial tool called the Absorbing Boundary Condition (CAP).

• What it is: In a computer simulation, the "universe" is finite. If an electron flies off, it hits the edge of the screen. Without a special rule, it would bounce back like a ball hitting a wall, creating fake noise. The CAP acts like a "black hole" or a sponge at the edge of the screen that swallows the electron so it doesn't bounce back.
• The Discovery: The researchers found that this "sponge" is actually a key part of the glitch.
  • When the sponge is on, it cleans up the "noise" of the simulation, leaving behind a very pure, simple wobble. The old math sees this pure wobble and accidentally amplifies it into the instability.
  • When the sponge is off, the simulation is "noisy" with many different frequencies interfering with each other. This messiness actually prevents the old math from finding that perfect rhythm to amplify, so the instability doesn't happen.

This proves the instability isn't a fundamental law of nature; it's a specific interaction between a "noisy" environment being cleaned up and a math formula that lacks memory.

Summary

• The Claim: The "dipole instability" (molecules suddenly wiggling wildly after a light pulse) reported in recent studies is not real. It is a mathematical artifact.
• The Cause: It is caused by using a simplified math method (adiabatic TDDFT) that lacks "memory," which accidentally amplifies tiny, natural vibrations into a runaway effect.
• The Proof: When the same simplified math is used in a more robust framework (RR-TDDFT) that separates space and time correctly, the instability vanishes.
• The Takeaway: Scientists should be careful when interpreting these specific types of computer simulations. Just because a computer says a molecule is going crazy doesn't mean the molecule is actually going crazy; it might just be the computer's math getting confused.

Technical Summary

Technical Summary: Post-pulse dipole instability in adiabatic TDDFT: fact or artifact?

Problem Statement
Recent real-time time-dependent density functional theory (TDDFT) simulations have reported an unexpected phenomenon: a delayed growth of molecular dipole oscillations occurring several femtoseconds after an extreme-ultraviolet (XUV) pulse is turned off. This "dipole instability" manifests as a resurgence of the dipole signal following a period of near-zero oscillation, growing with an exponential envelope before decaying. While previous studies suggested this might be a physical effect driven by population inversion and spontaneous symmetry breaking, the authors question whether this instability is a genuine physical prediction or an artifact inherent to the computational approximations used, specifically the adiabatic approximation within the traditional time-dependent Kohn-Sham (TDKS) framework.

Methodology
To resolve this ambiguity, the authors compare two distinct formulations of TDDFT for real-time non-equilibrium dynamics using the nitrogen ($N_2$) molecule as a test case:

1. Traditional TDKS: The standard approach solving the non-linear partial differential equations for Kohn-Sham orbitals (Eq. 1), where the exchange-correlation potential is approximated adiabatically (instantaneous dependence on density).
2. Response-Reformulated TDDFT (RR-TDDFT): A recently developed framework that propagates the coefficients of the many-body state expanded in a basis of field-free states via a set of ordinary differential equations (ODEs) (Eq. 2). Crucially, in RR-TDDFT, exchange-correlation quantities are evaluated only near ground-state or response densities, rather than on the fully non-equilibrium density.

The study employs the $N_2$ molecule subjected to off-resonant XUV pulses. Calculations utilize the Gaussian basis set cc-pVDZ (and aug-cc-pVDZ for convergence checks) and various exchange-correlation functionals, including PBE, PBE0, r2SCAN, and Time-Dependent Hartree-Fock (TDHF). To handle ionization continuum states, the authors implement Complex Absorbing Potentials (CAPs): a damping coefficient on molecular orbital energies for TDKS (via NWChem) and an analogous damping term on state amplitudes for RR-TDDFT. Results are benchmarked against approximate coupled-cluster (CC2) calculations where feasible.

Key Results

• Instability in TDKS: The authors successfully reproduce the post-pulse dipole instability in TDKS simulations across a range of pulse parameters, field strengths, and functional classes (GGA, hybrid, meta-GGA, and HF). The instability appears as a delayed exponential growth of dipole oscillations after the pulse ends.
• Absence in RR-TDDFT: Under identical conditions (same basis set, same functional, same pulse parameters, and analogous CAP treatment), the instability is completely absent in RR-TDDFT simulations. The dipole oscillations decay or remain stable without exponential growth.
• Role of the CAP: The instability is highly sensitive to the absorbing boundary condition. When the CAP is turned off in TDKS, the system retains a "noisy" spectrum of frequencies post-pulse, and the instability vanishes. The authors argue the CAP isolates specific resonant frequencies, allowing the adiabatic functional to drive them.
• Functional Independence: The phenomenon persists across different functionals (PBE, PBE0, r2SCAN, TDHF), though the onset time varies significantly (e.g., much slower for TDHF). This suggests the issue is not specific to self-interaction error in a particular functional but is a structural feature of the adiabatic approximation in the TDKS formalism.
• Comparison with CC2: RR-TDDFT results show closer agreement with CC2 reference data than TDKS results, particularly regarding the absence of unphysical growth.

Mechanism of the Artifact
The authors provide a numerical and analytical argument that the instability is an artifact of the incorrect non-linearity introduced by the adiabatic approximation in TDKS.

• After the pulse, the system is left with small, nearly pure sinusoidal oscillations at a specific frequency (corresponding to a linear-response excitation).
• In the adiabatic TDKS framework, the exchange-correlation potential depends instantaneously on the density. This creates a feedback loop where the potential term $v_{XC}$ resonates with the existing oscillation frequency.
• This "self-driving" effect causes the amplitude of the oscillations to grow exponentially until the density changes significantly enough to shift the resonant frequencies (a known issue of spurious peak-shifting in adiabatic TDDFT), at which point the growth peaks and decays.
• In contrast, RR-TDDFT evolves via linear ODEs (Eq. 2) in the absence of an external field. Without a field, the magnitude of the coefficients cannot grow exponentially; the system simply undergoes free evolution with exponential decay due to the CAP. The separation of space and time dependence in RR-TDDFT prevents the instantaneous density dependence from creating this artificial resonance.

Significance and Conclusions
The paper concludes that the post-pulse dipole instability observed in recent literature is an artifact of the adiabatic TDKS approach, not a physical phenomenon. The instability arises because the adiabatic approximation, when applied to the fully non-equilibrium density in TDKS, introduces a non-linearity that incorrectly amplifies small oscillations.

The authors highlight that RR-TDDFT offers a rigorous alternative that avoids this unphysical behavior while maintaining the computational efficiency of TDDFT. By evaluating exchange-correlation functionals only in the response regime (where they are derived), RR-TDDFT avoids the drastic approximation of applying ground-state functionals to far-from-equilibrium densities. This work underscores the importance of memory dependence (or the lack thereof in adiabatic approximations) in determining the stability of non-equilibrium dynamics and suggests that RR-TDDFT provides a more reliable framework for simulating ultrafast electron dynamics in molecules.