Simulating the exact real-time dynamics of a quantum many-body system is notoriously difficult. Because the state space grows exponentially with the number of particles, tracking the full wavefunction or density matrix quickly becomes computationally intractable.
To bypass this curse of dimensionality, physicists often look for a clever shorthand. Instead of tracking every possible correlation in the quantum state, what if we restrict our focus to a small, mathematically well-behaved family of states?
The Time-Dependent Variational Principle (TDVP) applied to Gaussian states provides exactly this shorthand. By constraining the quantum state to a Gaussian wavepacket or wave functional, the infinite-dimensional Schrödinger equation simplifies into a handful of coupled ordinary differential equations. This approach works because a Gaussian state is entirely characterized by its lowest-order statistical moments: its mean (the center of the state) and its covariance (its spread and internal quantum fluctuations).
The Geometry of Variational Projection
If you start with a Gaussian quantum state and let it evolve under a realistic, interacting Hamiltonian, it will immediately stop being Gaussian. Interactions introduce non-linearities that distort the state, stretching it into complex, non-Gaussian shapes.
The TDVP solves this problem geometrically. Imagine the space of all possible quantum states as a high-dimensional manifold, and the subset of all Gaussian states as a smooth, lower-dimensional surface (a sub-manifold) embedded within it. At any point on this Gaussian surface, the true quantum time evolution dictates a vector pointing off the surface, into non-Gaussian territory.
The TDVP takes this true evolutionary vector and projects it orthogonally onto the tangent space of the Gaussian surface. By doing so, it finds the "closest possible" Gaussian state to the true evolved state at every infinitesimal step in time.
As Berndt Müller and Tamás S. Bíró demonstrate in their work on lattice gauge fields, this variational projection successfully reduces the full Schrödinger equation to a set of ordinary differential equations governing the parameters of the wavepacket, such as its center and spread. Furthermore, because this projection respects the underlying geometric structure of quantum mechanics, the resulting equations of motion are not arbitrary approximations; as noted by Takeshi Sato et al., they inherently satisfy fundamental physical constraints like gauge invariance and the Ehrenfest theorem.
Moment-Based State Characterization
Why are Gaussian states such a powerful choice for this variational surface? The answer lies in their simplicity: they are completely determined by their first and second moments.
For a system of continuous variables, you do not need to track a complex wave function across a grid. You only need to track:
- First-order moments: The mean position and momentum vectors, which describe the classical-like trajectory of the system's center.
- Second-order moments: The covariance matrix, which describes the quantum uncertainty, squeezing, and entanglement between degrees of freedom.
In quadratic or linear systems, this representation is exact. Leehwa Yeh highlights that for quadratic systems, the mean vector and covariance matrix provide a complete analytical description of the state's evolution without ever requiring the full density matrix. When studying open systems—such as oscillators interacting with a thermal environment—Kuan-Liang Liu demonstrates that tracking the covariance matrix over time is sufficient to fully map out non-Markovian entanglement dynamics.
Even when the underlying physics is highly non-linear, treating the state as a Gaussian allows for exceptionally efficient control and optimization algorithms. For instance, Yu-Hong Liu et al. leveraged this moment-based reduction to design high-precision quantum learning control frameworks. Because the optimization algorithm only needs to compute updates for the first-order moments and the second-order covariance matrix, it completely sidesteps the massive computational overhead of full wave function simulations, enabling the discovery of optimal laser pulses for ultrafast cooling and entanglement generation.
Capturing Fluctuations and Back-Reaction
It is tempting to think that restricting a system to a Gaussian ansatz means discarding all the interesting quantum physics. However, the Gaussian variational framework is surprisingly rich because it explicitly models the interaction between the mean field and quantum fluctuations.
When a system undergoes a violent change—such as a phase transition or rapid particle production—the average field value (the first moment) and the quantum fluctuations around it (the second moment) couple together. As the mean field oscillates, it drives the creation of quantum fluctuations; conversely, those fluctuations exert a "back-reaction" that dampens or alters the mean field.
This mechanism is beautifully illustrated in the context of nonequilibrium field theories. Fred Cooper et al. showed that approximating a quantum system's state with a Gaussian density matrix allows researchers to simultaneously track the mean field and two-point correlation functions, capturing the critical leading-order dynamics of quantum fluctuations during symmetry breaking.
Even though the underlying equations remain fundamentally reversible and Hamiltonian, this coupling between the mean field and its fluctuations can give rise to "apparent" irreversibility. Salman Habib et al. demonstrated that the dephasing of these quantum fluctuations mimics the effects of dissipation and quantum decoherence without requiring any actual loss of information from the total system.
+-------------------------------------------------------+
| True Quantum Evolution |
| (Infinite-dimensional, highly complex state space) |
+-------------------------------------------------------+
|
| TDVP Orthogonal
| Projection
v
+-------------------------------------------------------+
| Gaussian Variational Manifold |
| Characterized entirely by low-dimensional parameters |
+-------------------------------------------------------+
/ \
v v
+-----------------------+ +-----------------------+
| First-Order Moments |<------------->| Second-Order Moments |
| (Mean Field / Center) | Coupled Loops | (Covariance Matrix / |
+-----------------------+ | Quantum Fluctuations) |
+-----------------------+
This capability makes the Gaussian variational ansatz a versatile tool across diverse areas of physics:
- Bose-Einstein Condensates (BECs): Kai Huang et al. established that the Gaussian variational method is mathematically equivalent to the self-consistent Bogoliubov field method, making it an elegant framework for evaluating time-dependent phenomena like the breathing modes of trapped atomic gases.
- Lattice Gauge Theories: By applying clever transformations to decouple gauge fields from matter, Pablo Sala et al. used the TDVP to evolve the covariance matrix of a Gaussian ansatz, successfully capturing complex, real-time dynamical processes like string breaking.
- Phase-Space Solitons: K. Baudin et al. utilized a Gaussian ansatz to reduce wave packet dynamics to the motion of point-like particles in a potential well, revealing how coherent soliton states can remain stable and trapped within a sea of random gravitational noise.
Stochastic Dynamics and Continuous Measurement
The utility of the Gaussian approximation extends beyond closed, isolated systems. When a quantum system is subjected to continuous monitoring or environmental noise, its state changes stochastically. Remarkably, if the initial state is Gaussian and the measurement processes are linear, the state remains Gaussian, allowing the same moment-based intuition to hold.
Instead of solving a master equation for a full density matrix, the system can be modeled using coupled stochastic differential equations for the moments. For example, in the field of Magnetic Resonance Force Microscopy (MRFM), Shesha Raghunathan et al. treated a mechanical cantilever's quantum state as a Gaussian wavepacket. This reduction compressed a complex density matrix into just 11 coupled stochastic equations that accurately tracked the mean position and momentum variances, drastically lowering the computational barrier required to simulate single-spin measurements.
When continuous measurement is used to actively prepare or stabilize a specific quantum state, the evolution of the second-order moments decouples from the measurement outcomes in a unique way. As Liying Bao et al. showed, the conditional covariance matrix in a continuous measurement framework is governed by a deterministic Riccati equation. Because this equation is deterministic, researchers can precisely analyze how continuous measurements shrink the covariance matrix, allowing them to prepare highly squeezed, stable target states more efficiently than through unconditional, dissipation-only dynamics.
By trading the completeness of the full quantum state space for the targeted precision of the first and second moments, the time-dependent variational principle for Gaussian states turns intractable quantum problems into elegant, solvable dynamical systems.
Key Foundational Work
Methods for a nonuniform Bose gas (1996) — contributed 2 concept elements to this direction.
Variational approach to real-time evolution of Yang-Mills gauge fields on the lattice (1994) — contributed 1 concept elements to this direction.
Steady states of harmonic oscillator chains and shortcomings of harmonic heat baths (1994) — contributed 1 concept elements to this direction.