Skip to content

Multi-Configuration Time-Dependent Hartree

A variational quantum-dynamics method that expands a multidimensional wavefunction in configurations built from time-dependent single-particle functions and evolves both coefficients and basis functions.

Version
v2 · 2026-09-06 · History
Domain-specific #
2322
Origin domain
quantum dynamics
Subdomain
wavepacket propagation
Aliases
MCTDH, Multiconfigurational time-dependent Hartree

Core Idea

Multi-Configuration Time-Dependent Hartree (MCTDH) approximates a multidimensional time-dependent wavefunction with a sum of Hartree products whose one-coordinate factors move with time:

\[ \Psi(q_1,\ldots,q_f,t)= \sum_{j_1=1}^{n_1}\cdots\sum_{j_f=1}^{n_f} A_{j_1\cdots j_f}(t) \prod_{\kappa=1}^{f} \varphi^{(\kappa)}_{j_\kappa}(q_\kappa,t). \]

Both the configuration coefficients \(A_J(t)\) and the single-particle functions are determined variationally. A fixed primitive product basis can require exponentially many coefficients; adapting the one-dimensional bases to the evolving wavepacket can represent correlated dynamics with fewer configurations. Meyer, Manthe, and Cederbaum introduced the method in 1990.[1]

The identity is the coupled time-dependent variational optimization of coefficients and basis functions. A static configuration-interaction expansion or a single Hartree product lacks this dual adaptation.

Structural Signature

  • Time-dependent Schrödinger problem: a Hamiltonian acts on multiple coupled coordinates.
  • Mode partition: physical coordinates are assigned to modes \(q_\kappa\).
  • Primitive bases: each mode has a numerical grid or fixed basis supporting its functions.
  • Time-dependent SPFs: a small orthonormal set \(\varphi_j^{(\kappa)}(t)\) spans each adaptive modal subspace.
  • Configuration tensor: coefficients \(A_J(t)\) combine SPF products.
  • Variational principle: the Dirac–Frenkel condition projects evolution onto the ansatz manifold.
  • Coupled equations: coefficients and SPFs evolve together through mean-field operators and density matrices.
  • Gauge constraint: orthonormality and tangent-space conventions remove representation redundancy.
  • Convergence hierarchy: increasing SPF counts approaches the primitive-basis solution.
  • Hamiltonian representation: efficient propagation commonly relies on sum-of-products operator form.

Recognition test. Confirm that the expansion contains multiple configurations and time-dependent modal basis functions, and that both layers are propagated variationally. “Time-dependent Hartree” with one product and multiconfiguration propagation in a fixed orbital basis are neighboring methods.

What It Is Not

MCTDH is not classical molecular dynamics: it propagates a quantum wavefunction, not trajectories on a single potential surface. It is not mixed quantum–classical dynamics, where selected degrees of freedom are classical.

It is not ordinary time-dependent Hartree. One configuration cannot generally represent intermode correlation. It is not time-dependent configuration interaction in a fixed basis, because adaptive SPFs are load-bearing.

Standard MCTDH for distinguishable modes is also not automatically MCTDHF or MCTDHB, which impose fermionic determinants or bosonic permanents for indistinguishable particles. Multilayer MCTDH recursively factors tensors and is an extension, not the base identity.

Scope of Application

MCTDH is used for molecular wavepacket dynamics, photodissociation, reactive scattering, vibrational energy redistribution, nonadiabatic dynamics, and model open-system Hamiltonians.[2] It is especially effective when a moderate number of modes have structured coupling and the wavefunction occupies a low-rank adaptive subspace.[3]

The method requires choices: coordinate system, mode combination, primitive grids, SPF counts, integrator, and Hamiltonian representation. A poor partition can leave the coefficient tensor exponentially large. Mode combination groups coordinates into multidimensional SPFs to balance correlation and basis cost.

Potential-energy and kinetic operators are often expressed in sums of products. General potential fitting and multilayer variants broaden reach, but numerical tractability is not guaranteed by the name alone.

Convergence must be checked along more than one axis. Enlarging a primitive grid cannot compensate for too few SPFs, while enlarging SPF sets cannot repair a truncated coordinate range or inaccurate Hamiltonian. Time-step convergence, norm and energy behavior where applicable, natural-population decay, and observable stability under basis enlargement provide complementary diagnostics. A single visually smooth wavepacket is not a convergence certificate.

Clarity

The coefficient tensor captures correlation among modes relative to the current SPFs. The SPFs themselves rotate to follow occupied directions. These adaptations must be separated: more configurations expand correlation rank, while better SPFs improve the moving modal subspaces.

For orthonormal SPFs, a common gauge is

\[ \left\langle\varphi_j^{(\kappa)} \middle| \dot\varphi_l^{(\kappa)} \right\rangle=0. \]

Without a gauge, rotations among SPFs could be offset by inverse rotations of \(A\), leaving \(\Psi\) unchanged and making equations nonunique.

“Single-particle function” is a modal factor name; the modes need not correspond to literal particles.

Manages Complexity

A direct product basis with \(N\) functions for each of \(f\) modes contains \(N^f\) coefficients. MCTDH replaces full primitive occupancy with \(n_\kappa\) adaptive SPFs, yielding \(\prod_\kappa n_\kappa\) configurations plus SPF coefficients. When \(n_\kappa\ll N_\kappa\), this is a major reduction.

The method moves approximation effort where the wavefunction actually travels. Natural populations—eigenvalues of modal density matrices—diagnose whether omitted SPFs matter. Small populations at the truncation boundary support, but do not alone prove, convergence.

Costs remain exponential in the number of modes at the coefficient layer. Multilayer tensor trees address this remaining bottleneck rather than negating the original method.

Abstract Reasoning

Let the trial manifold consist of normalized ansatz states \(\Psi(z)\) parameterized by \(A\) and the SPFs. Dirac–Frenkel variation requires

\[ \left\langle\delta\Psi\middle| i\frac{\partial}{\partial t}-H \middle|\Psi\right\rangle=0 \]

for all tangent variations. Variations in \(A^\ast\) yield a Schrödinger-like coefficient equation in the SPF product basis. SPF variations yield projected mean-field equations involving reduced density matrices.[3]

If every SPF space equals its full primitive space, the ansatz spans the primitive direct product and propagation becomes exact within that discretization. If each mode has one SPF, the ansatz reduces to time-dependent Hartree. This gives a controlled hierarchy between mean field and full product propagation.

The hierarchy is controlled by representation size, but convergence is problem-dependent. Increasing the number of single-particle functions enlarges the variational manifold; it does not say how quickly an observable converges or eliminate numerical errors from primitive grids, time propagation, regularization, or reduced-density inversion. Reference-grade use therefore reports both the configuration dimensions and the primitive discretization, then checks observables under systematic enlargement rather than treating the method's variational derivation as an error certificate.

Knowledge Transfer

The adaptive low-rank idea transfers to tensor trains, time-dependent variational matrix-product states, and multilayer tensor networks: evolve a state on a structured low-rank manifold rather than in the full tensor product. The precise MCTDH identity remains the sum of Hartree products with modal SPFs and its variational equations.

Transfer fails when one merely changes basis after propagation or truncates singular values without coupled variational time evolution. Similar numerical economy does not make every tensor method MCTDH.

Examples

  1. Coupled oscillators. The original paper showed convergence for a multidimensional coupled-oscillator model as configurations increased.[1]
  2. Photodissociation. An excited wavepacket moves and splits across coupled nuclear coordinates while SPFs follow occupied regions.
  3. Nonadiabatic dynamics. Electronic-state components can be combined with nuclear modes in a multiset formulation.
  4. One SPF per mode. The representation collapses to a Hartree product and misses general intermode entanglement.
  5. Full SPF sets. The method reaches exact primitive-basis propagation, subject to numerical integration error.
  6. Multilayer extension. Recursively factorizing the coefficient tensor treats more modes but changes the ansatz topology.

Structural Tensions

  • Adaptive efficiency vs. nonlinear equations: moving bases reduce rank but couple propagation equations. Diagnostic: monitor integrator error and density conditioning.
  • Compactness vs. correlation: too few SPFs suppress entanglement. Diagnostic: increase SPF counts and inspect natural populations.
  • Mode choice vs. cost: physically intuitive coordinates can produce inefficient correlations. Diagnostic: compare mode partitions.
  • Variational exactness vs. operator fitting: ansatz propagation may be variational while a fitted Hamiltonian is approximate. Diagnostic: audit potential representation separately.
  • Gauge freedom vs. numerical stability: redundant basis rotations can destabilize equations. Diagnostic: state and enforce the gauge.
  • Autonomy vs. generic Approximation: MCTDH adds a precise moving-basis tensor manifold and equations. Diagnostic: require both coefficient and SPF evolution.

Structural–Framed Character

The abstraction is structural in its adaptive tensor factorization and tangent-space evolution, but framed by quantum mechanics. Wavefunctions, Hamiltonians, Schrödinger time, and modal products are literal.

It is autonomous because the same ansatz and variational machinery recurs across molecular systems, not because it is a particular software package.

Structural Core vs. Domain Accent

The core is time-dependent variational low-rank approximation. The domain accent is a quantum wavefunction expanded in Hartree products of adaptive SPFs. Removing quantum roles yields a broader tensor method; freezing SPFs yields configuration interaction.

Its applications occupy one quantum-dynamics lineage, so it is domain-specific.

Approximation is the proposed minimal parent: MCTDH is a strict, controlled approximation by an adaptive variational manifold. Decomposition describes the product expansion, and Molecular Dynamics is a neighboring simulation family, but neither is its genus.

Relationships to Other Abstractions

Local relationship map for Multi-Configuration Time-Dependent HartreeParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Multi-Configuration …DOMAINPrime abstraction: Approximation — is a kind ofApproximationPRIME

Current abstraction Multi-Configuration Time-Dependent Hartree Domain-specific

Parents (1) — more general patterns this builds on

  • Multi-Configuration Time-Dependent Hartree is a kind of Approximation Prime

    Approximation is the proposed minimal parent: MCTDH is a strict, controlled approximation by an adaptive variational manifold.

Hierarchy path (1) — routes to 1 parentless root

Neighborhood in Abstraction Space

Multi-Configuration Time-Dependent Hartree sits in a sparse region of the domain-specific corpus (93rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With

  • Time-dependent Hartree: one configuration.
  • Time-dependent configuration interaction: multiple fixed-basis configurations.
  • MCTDHF/MCTDHB: symmetry-adapted fermionic or bosonic methods.
  • Multilayer MCTDH: recursive extension.
  • Mixed quantum–classical dynamics: propagates some variables classically.
  • Molecular dynamics: normally classical nuclear trajectories.

References

[1] H.-D. Meyer, U. Manthe, and L. S. Cederbaum, “The Multi-Configurational Time-Dependent Hartree Approach,” Chemical Physics Letters 165, no. 1 (1990): 73–78, https://doi.org/10.1016/0009-2614(90)87014-I. registry ↩a ↩b

[2] H.-D. Meyer, “Studying Molecular Quantum Dynamics with the Multiconfiguration Time-Dependent Hartree Method,” WIREs Computational Molecular Science 2 (2012): 351–374, https://doi.org/10.1002/wcms.87. registry ↩

[3] M. H. Beck, A. Jäckle, G. A. Worth, and H.-D. Meyer, “The Multiconfiguration Time-Dependent Hartree Method: A Highly Efficient Algorithm for Propagating Wavepackets,” Physics Reports 324 (2000): 1–105, https://doi.org/10.1016/S0370-1573(99)00047-2. registry ↩a ↩b