schrödinger equation
The Schrödinger equation governs the coherent evolution of nonrelativistic quantum states and, in its stationary form, defines the energy eigenstates of a specified Hamiltonian. It predicts complex probability amplitudes rather than classical trajectories or direct measurement outcomes. A complete problem must declare the Hilbert space, Hamiltonian and operator domain, particle statistics, boundary and initial conditions, potentials and fields, normalization, approximation regime, and observable model. In semiconductor devices these choices control confinement, tunneling, subbands, wavepacket motion, transport, spin, valleys, optical transitions, and self-consistent charge.
```svg
```
**The time-dependent Schrödinger equation is a first-order evolution law.** $i\hbar\partial_t|\psi(t)\rangle=\hat H(t)|\psi(t)\rangle$ specifies how a prepared state evolves between measurements. First order in time means one initial state is required, unlike the position and velocity data of a classical second-order equation. The Hamiltonian may be time dependent through drives, moving boundaries, or changing fields.
**The position representation turns operator evolution into a complex partial differential equation.** For one spinless particle with scalar mass and potential, $i\hbar\partial_t\psi(\mathbf r,t)=[-\hbar^2\nabla^2/(2m)+V(\mathbf r,t)]\psi(\mathbf r,t)$. The Laplacian supplies dispersion and the potential supplies phase and force structure. Spin, magnetic fields, heterogeneous mass, relativity, and interactions require additional terms or components.
**The wavefunction is a probability amplitude rather than a material wave density.** $|\psi(\mathbf r,t)|^2$ gives position probability density under the Born rule for a normalized pure state. Complex phase does not appear in density alone but controls interference and current. Multiplying the entire state by one global phase changes no observable; spatially varying or relative phase is physically consequential.
**Normalization fixes total probability for a bound-state wavefunction.** Require $\int|\psi|^2d^3r=1$ for a single-particle pure state over the modeled domain. Plane waves and scattering eigenstates are generalized states normalized to delta functions or flux, not ordinary square-integrable vectors. Finite-box normalization is a computational convention whose volume factors must cancel from physical observables.
**Self-adjoint Hamiltonians generate unitary closed-system evolution.** With a suitable operator domain, the Hamiltonian gives a norm-preserving propagator. Formal Hermiticity of the differential expression is insufficient if boundary terms do not vanish or interface conditions violate current conservation. An absorbing boundary intentionally breaks unitarity in the retained region and should be labeled as an open-boundary approximation.
**Probability current turns norm conservation into a local continuity law.** For a scalar potential and constant mass, $\rho=|\psi|^2$ and $\mathbf j=(\hbar/m)\operatorname{Im}(\psi^*\nabla\psi)$ satisfy $\partial_t\rho+\nabla\cdot\mathbf j=0$. Current through a boundary changes enclosed probability. Vector potentials and multiband Hamiltonians modify the current operator; reusing the scalar formula can violate conservation.
**Boundary conditions are part of the Hamiltonian domain.** Dirichlet, Neumann, Robin, periodic, interface, outgoing, and absorbing conditions represent different physical problems. An infinite wall imposes zero amplitude, while a finite barrier requires matching consistent with the kinetic operator. Artificial domain boundaries must be far enough away or treated to prevent reflected waves from contaminating the observable.
**Initial conditions must belong to the state space and operator regime being evolved.** A normalized square-integrable packet can evolve even if it is not an energy eigenstate. Discontinuous trial states may have infinite kinetic-energy expectation and stress numerical grids. A state prepared by a physical source has finite bandwidth, spatial extent, spin, and phase uncertainty that should be included rather than assumed away.
```svg
```
**Separation of variables produces the stationary equation only for suitable time dependence.** When $H$ is time independent, solutions can be expanded in states $\psi_n(\mathbf r)e^{-iE_nt/\hbar}$ satisfying $H\psi_n=E_n\psi_n$. The time-independent Schrödinger equation is an eigenvalue problem, not a separate universal dynamics law. A general state is a superposition of stationary components.
**Energy eigenstates are stationary in probability but still accumulate phase.** A nondegenerate eigenstate changes by a global phase, leaving fixed-position density and time-independent expectation values of fixed observables. Superpositions of unequal energies develop relative phase and can show beating. Degenerate superpositions can remain stationary under the unperturbed Hamiltonian while perturbations select new combinations.
**The energy spectrum can be discrete, continuous, or mixed.** Confining potentials often yield bound discrete levels; open motion yields continuous scattering energies; realistic potentials can have both. Resonances are metastable scattering structures rather than normalizable bound eigenstates. A finite numerical box discretizes the continuum, so mesh eigenvalues above threshold are not automatically device levels.
**Expectation values follow from operators and the evolving state.** $\langle A\rangle=\langle\psi|\hat A|\psi\rangle$ is an ensemble average, not necessarily an individual measurement outcome. Time evolution can be assigned to states, operators, or both through equivalent pictures. Measurement apparatus, projectors or POVMs, and preparation complete the prediction beyond the differential equation.
**Ehrenfest’s theorem connects quantum averages to classical-looking equations.** For $H=p^2/(2m)+V(x)$, $d\langle x\rangle/dt=\langle p\rangle/m$ and $d\langle p\rangle/dt=-\langle V'(x)\rangle$. This is not generally $-V'(\langle x\rangle)$ unless the potential is at most quadratic or the packet is sufficiently narrow. Wavepacket spread and interference preserve genuinely quantum behavior.
**Free-particle wavepackets disperse because energy is nonlinear in momentum.** Each momentum component accumulates phase $e^{-i\hbar k^2t/(2m)}$, causing a Gaussian packet to broaden while its center moves at group velocity. A plane wave has definite momentum but infinite extent and cannot represent localized preparation. Dispersion differs from environmental decoherence: a pure state can spread unitarily.
**Fourier transformation interchanges position and momentum descriptions.** The position wavefunction and momentum amplitude are Fourier pairs under normalization conventions. The kinetic operator is diagonal in momentum space, while a local potential is diagonal in position space. Split-operator methods exploit this complementarity. Grid spacing and domain length impose reciprocal cutoffs that must cover the packet spectrum.
**The uncertainty relation reflects noncommuting operators and state geometry.** $\Delta x\Delta p\ge\hbar/2$ follows from commutation and Cauchy–Schwarz. It is not caused by a particular measurement instrument alone. Gaussian minimum-uncertainty states saturate the bound under conditions. A narrow spatial grid representation requires broad momentum support; truncation can violate the intended state.
**The infinite square well makes boundary quantization explicit.** Zero wavefunction at two walls permits standing waves with discrete $E_n\propto n^2/L^2$. The infinite potential is an ideal limit, not a semiconductor band offset. Finite barriers lower energies relative to the infinite model and allow evanescent leakage. Centering the well changes parity convenience but not physical spectrum.
**The finite square well separates bound, evanescent, and continuum behavior.** Bound energies satisfy transcendental matching conditions and wavefunctions decay outside. Only finitely many bound states exist for fixed depth and width. Near-threshold states extend far beyond the nominal well and are sensitive to domain truncation. Effective mass discontinuity modifies derivative matching at heterointerfaces.
```svg
```
**A delta potential exposes matching conditions and dimensional coupling.** An attractive one-dimensional delta well supports one bound state, while the derivative jumps according to integrated Schrödinger equation. The wavefunction remains continuous under the standard model. Delta interactions idealize short-range features and require regularization or renormalization in higher dimensions. Their simplicity makes them useful verification cases.
**The harmonic oscillator combines confinement with exact ladder structure.** A quadratic potential yields equally spaced levels $E_n=\hbar\omega(n+1/2)$ and Hermite–Gaussian eigenfunctions. The ground state has zero-point energy and minimum uncertainty. Coherent states move with classical center motion without shape change. Anharmonic device potentials break equal spacing and generate amplitude-dependent transitions.
**Central potentials reduce three-dimensional motion through angular momentum.** Separation in spherical coordinates gives spherical harmonics and a radial equation with centrifugal effective potential. Regularity at the origin and square integrability constrain solutions. Orbital quantum numbers arise from rotation symmetry. Crystal fields and device boundaries break spherical symmetry and mix angular sectors.
**The hydrogen atom demonstrates Coulomb spectrum and degeneracy.** Its nonrelativistic Schrödinger solution gives bound energies scaling as $-1/n^2$ and continuum ionization states. Degeneracies reflect rotation and hidden symmetry. Fine structure, Lamb shift, nuclear size, spin, and relativistic effects lie beyond the basic equation. Semiconductor hydrogenic dopants use dielectric screening and effective mass, not vacuum constants.
**One-dimensional node theorems order bound states by zeros.** For regular Sturm–Liouville-like potentials, the ground state has no interior node and excited states gain nodes in energy order. This helps identify numerical eigenpairs and sketch qualitative solutions. Multidimensional nodal geometry is more complex, while degeneracy can undermine simple ordering. Spurious grid oscillations should not be mistaken for physical nodes.
**Classically forbidden regions support exponential amplitude rather than zero probability.** Where $V>E$ for a stationary scalar problem, local solutions grow or decay exponentially. Physical boundaries select combinations. Finite penetration shifts bound energies and permits tunneling. “Forbidden” refers to classical kinetic-energy sign, not impossibility in quantum mechanics.
**Barrier tunneling depends exponentially on action through the forbidden region.** Transmission falls approximately as $\exp[-2\int\kappa(x)dx]$ in a WKB regime with $\kappa=\sqrt{2m(V-E)}/\hbar$. Prefactors, turning points, resonances, dimensionality, and effective mass matter. Exponential sensitivity makes barrier thickness, height, and field uncertainty decisive in gate leakage and tunnel junctions.
**Resonant tunneling uses interference between multiple barriers.** Quasibound states in a well align with incident energy and enhance transmission toward unity in ideal coherent symmetric structures. Contact coupling sets resonance width and lifetime. Bias shifts the potential self-consistently, while scattering and temperature broaden response. A stationary closed-well eigenvalue alone cannot predict current.
**The WKB approximation separates slowly varying phase and amplitude.** It is valid when the local wavelength changes slowly away from turning points. Connection formulas bridge oscillatory and evanescent regions. WKB estimates quantization, tunneling, and semiclassical propagation but fails near abrupt features, low quantum numbers, interference caustics, or closely spaced turning points without uniform corrections.
**The variational method bounds the ground-state energy from above.** A normalized trial wavefunction in the Hamiltonian domain gives $\langle H\rangle\ge E_0$. Optimizing parameters improves the bound. Energy can appear accurate while tails, nodes, transition matrix elements, or interface density remain poor. Excited-state bounds require orthogonality or subspace methods.
**Perturbation theory expands around a solvable stationary equation.** With $H=H_0+\lambda V$, energy and state corrections involve unperturbed matrix elements and level gaps. Near degeneracy, first diagonalize within the degenerate subspace. Small potential amplitude alone is insufficient if gaps are smaller. Stark, Zeeman, strain, and interface perturbations illustrate the method.
**Time-dependent perturbations drive transitions through spectral overlap.** In the interaction picture, coupling matrix elements and oscillatory phases determine amplitudes. Near resonance, a periodic drive can produce Rabi oscillations; weak continuum coupling yields Fermi’s golden rule under long-time assumptions. Pulse envelope, bandwidth, selection rules, decoherence, and extra levels determine experimental response.
**The adiabatic approximation follows instantaneous eigenstates only with adequate gaps and slow change.** A slowly varying potential can transport a state while accumulating dynamic and geometric phase. Small avoided crossings or rapid endpoints cause transitions. Device ramps should be assessed through coupling matrix elements divided by gap scales, not ramp duration alone. Disorder can introduce unexpected small gaps.
**The imaginary-time equation projects toward low-energy states.** Replacing real time by $-i\tau$ turns unitary phase evolution into exponential energy filtering. Repeated normalization suppresses excited components when the initial state overlaps the ground state. The method is computational, not physical real-time dynamics. Excited states require orthogonality or block methods, and stiffness can demand implicit schemes.
**Spinor Schrödinger equations couple spatial amplitudes to internal states.** Pauli spin terms, Zeeman coupling, spin–orbit interaction, valley, band, and sublattice degrees produce multicomponent wavefunctions and matrix differential operators. Probability current and boundary conditions must be derived from the full Hamiltonian. Component norms are basis dependent, while total observables are not.
**Magnetic fields require gauge-covariant kinetic momentum.** Minimal coupling uses $-i\hbar\nabla-q\mathbf A$ and scalar potential $q\phi$. Gauge transformations change potentials and wavefunction phase while preserving density and current. A discrete grid must encode link phases or compatible covariant derivatives to avoid gauge-dependent spectra. Landau levels emerge for uniform fields.
**Identical particles lift the equation into configuration space.** An $N$-particle wavefunction depends on $3N$ spatial coordinates plus internal labels and must be symmetric for bosons or antisymmetric for fermions. Interaction terms couple coordinates, making direct solution exponentially difficult. Mean-field, density-functional, configuration-interaction, tensor-network, and Monte Carlo methods reduce or approximate the problem differently.
**The Schrödinger equation has a defined nonrelativistic domain of validity.** It does not create or destroy particles, include relativistic covariance, or automatically include spin. The Pauli, Dirac, Klein–Gordon, and quantum-field equations cover other regimes. Effective Schrödinger-like equations remain useful in solids because quasiparticles have low-energy dispersions and parameters different from free vacuum particles.
```svg
```
**Finite differences replace derivatives with local grid stencils.** Central differences produce sparse kinetic matrices and converge with order determined by stencil and smoothness. Grid spacing must resolve the shortest wavelength and interface variation. Abrupt mass changes require flux-consistent discretization. Boundary rows are part of the operator and can destroy Hermiticity if assembled inconsistently.
**Finite elements use a weak Schrödinger eigenproblem on flexible geometry.** Basis functions and quadrature yield Hamiltonian and overlap matrices $Hc=ESc$. The mass or overlap matrix defines normalization and orthogonality. Mesh refinement can target interfaces, corners, and wells. Spurious modes, poor elements, quadrature, and artificial boundaries need convergence tests.
**Spectral methods expand the wavefunction in global basis functions.** Fourier, oscillator, spherical harmonic, plane-wave, and problem-adapted bases can converge rapidly for smooth solutions. Discontinuities and localized interfaces slow convergence or cause ringing. Basis cutoffs define ultraviolet resolution and must cover driven or tunneling states, not only the ground state.
Matrix eigensolvers should target the relevant spectral region. Dense diagonalization scales poorly; Lanczos, Arnoldi, and shift-invert methods compute selected eigenpairs of sparse operators. Residual norm, orthogonality, and basis convergence accompany each level. Near degeneracy, compare projectors or subspaces rather than eigenvector signs and ordering.
The shooting method integrates a one-dimensional stationary equation while varying energy until boundary conditions match. Node count brackets bound states and log derivatives improve stability. Exponentially growing unwanted solutions can dominate long forbidden regions. Multiple wells, near degeneracy, and discontinuous mass favor matching or matrix methods.
Transfer matrices connect amplitudes across layered one-dimensional regions but can become ill-conditioned when growing and decaying exponentials coexist. Scattering matrices and recursive Green functions are more stable for thick barriers or many layers. Interface ordering and flux normalization must be consistent. Determinant drift can reveal numerical failure.
**Crank–Nicolson gives a norm-preserving second-order update for time-independent Hermitian discretizations.** The centered implicit step is unitary in the discrete metric when solved accurately. It requires a linear solve each step and can retain unresolved high-frequency oscillations rather than damp them. Time dependence needs careful midpoint evaluation; nonlinear self-consistency adds iteration error.
Explicit Euler is unstable for standard unitary Schrödinger evolution because amplification increases norm. Implicit Euler damps and is not unitary. General Runge–Kutta methods can be accurate over short times but require norm, phase, and stability checks. Renormalizing after each step hides systematic nonunitarity and changes nonlinear or open-system physics.
Split-operator propagation alternates exponentials of kinetic and potential terms, often using FFTs. Strang splitting is second order and unitary for real potentials with exact substeps. Error arises from noncommutation and depends on gradients and timestep. Magnetic fields, position-dependent mass, nonlinear potentials, and complex boundaries weaken the simple separable split.
Krylov propagation approximates $e^{-iH\Delta t/\hbar}\psi$ in a state-dependent subspace. It handles sparse nonseparable Hamiltonians and can estimate local exponential error. Krylov dimension, timestep, reorthogonalization, and matrix norm affect accuracy. For time-dependent $H$, midpoint freezing or Magnus–Krylov schemes introduce ordering approximations.
Chebyshev propagation expands the exponential in stable polynomials after scaling the Hamiltonian spectrum to a bounded interval. It can achieve high accuracy for long time-independent steps. Incorrect spectral bounds cause divergence, while overly broad bounds waste terms. Time-dependent or non-Hermitian problems require modified approaches.
```svg
```
**Absorbing boundaries trade exact unitarity for an open-domain approximation.** Complex absorbing potentials, mask functions, exterior complex scaling, perfectly matched formulations, and transparent boundary kernels suppress reflection differently. Absorption should begin where physical interaction is negligible and vary smoothly relative to wavelength. Test reflection across energy and angle, not only one packet.
Open leads can instead be represented by scattering boundary conditions or energy-dependent self-energies. This turns the stationary device problem into a Green-function or nonlinear-energy effective operator. Lead modes require flux normalization. Artificial broadening should be distinguished from physical contact coupling and inelastic scattering.
Probability-current conservation is a stringent discretization test. Sum fluxes through all boundaries and compare with norm change or source terms. Local current should be derived from the discrete Hamiltonian, especially for tight binding, variable mass, and magnetic phases. A visually smooth density can coexist with a nonconservative current.
**Nonlinear Schrödinger equations are related models with different physics.** Mean-field interactions can add terms such as $g|\psi|^2\psi$ for Bose condensates or nonlinear optics. Superposition no longer holds and normalization can couple to parameters. The Gross–Pitaevskii equation, nonlinear envelope equations, and Kohn–Sham equations should not be confused with the linear single-particle Schrödinger equation.
Kohn–Sham equations are self-consistent effective one-particle eigenproblems from density-functional theory. Their potential depends on total density through Hartree and exchange-correlation terms. Kohn–Sham eigenvalues are not universally quasiparticle energies, though selected ones have interpretations. Basis, functional, pseudopotential, and convergence affect materials predictions.
Stochastic Schrödinger equations unravel certain master equations into ensembles of random pure-state trajectories. Individual trajectories depend on unraveling and can represent conditional measurement records or computational devices. Ensemble density operators carry invariant predictions. They do not mean an isolated system has classical random force unless the physical model specifies it.
The Lindblad master equation evolves density matrices, not wavefunctions, for Markovian open systems. A non-Hermitian effective Hamiltonian plus random quantum jumps is one unraveling. Relaxation and dephasing require jump operators and rates beyond the closed Hamiltonian. Using an imaginary potential alone cannot reproduce arbitrary decoherence.
**Poisson–Schrödinger coupling makes semiconductor confinement self-consistent.** Quantum states determine occupied carrier density; density enters Poisson’s equation; electrostatic potential returns to the Schrödinger Hamiltonian. Gate work functions, dopants, fixed charge, dielectric interfaces, temperature, Fermi level, exchange-correlation, and degeneracy close the model. Mixing or Newton methods solve the nonlinear loop.
Occupation is not determined by bound energies alone. Fermi–Dirac statistics, contact chemical potentials, dimensional density of states, spin and valley degeneracy, and nonequilibrium injection determine populations. Summing normalized probability densities without occupations gives the wrong charge. Open transport requires lesser Green functions or scattering-state filling rather than equilibrium subband rules.
Effective-mass Schrödinger equations replace vacuum electron mass with band-curvature parameters. Anisotropic valleys use mass tensors; nonparabolicity makes mass energy dependent or demands multiband models. At heterointerfaces, a symmetric flux-conserving kinetic operator and matching condition should be chosen. Parameter sets must match crystal orientation, strain, temperature, and band edge.
**Quantum wells turn layer stacks into subband eigenproblems.** Band offsets define finite confinement, material masses affect kinetic energy, and fields tilt the profile. Wavefunction penetration influences optical overlap and tunneling. Interface roughness and alloy disorder broaden and mix subbands. Spectroscopy validates transition differences and matrix elements, not an arbitrary absolute potential zero.
In inversion layers and nanowires, confinement redistributes charge away from a classical interface sheet and raises subband energies. This changes capacitance, threshold, density of states, and scattering. One-dimensional confinement slices coupled to semiclassical transport are efficient when longitudinal variation is slow. Full multidimensional quantum transport is needed when mode mixing and tunneling dominate.
Silicon device equations require valley structure beyond one scalar band. Different valleys have anisotropic masses and orientation-dependent confinement energy. Interface steps, electric field, strain, and atomic-scale disorder mix valleys and set valley splitting. A smooth effective-mass equation may need calibrated boundary or coupling terms from atomistic models.
Multiband $k\cdot p$ Schrödinger equations use spinor envelope functions and matrix differential operators to capture conduction–valence coupling, heavy and light holes, split-off bands, spin, and nonparabolicity. Operator ordering and interface conditions are model choices. Spurious solutions can appear if parameters or basis truncation violate the model’s validity range.
```svg
```
**Tunnel-current prediction needs contacts and occupation beyond a closed eigenproblem.** WKB can estimate leakage through a slowly varying barrier; transfer matrices handle coherent layers; NEGF handles open reservoirs and self-consistency; master equations handle selected incoherent regimes. Choosing by convenience can miss resonance, scattering, or charging. The measured current also includes area, temperature, series resistance, and defects.
Scanning tunneling microscopy relates current exponentially to tip–sample separation and local electronic states under approximations. The wavefunctions satisfy vacuum-barrier Schrödinger behavior, but measured topography convolves density of states, tip shape, bias, and feedback. An apparent height is not purely geometric. Atomic-scale interpretation often uses Tersoff–Hamann or more detailed tunneling models.
Electron microscopy uses relativistically corrected wavelength and electron-optical propagation, while elastic specimen scattering can be formulated through stationary or paraxial Schrödinger-like equations. Multislice propagation alternates transmission and free-space steps. Inelastic scattering, partial coherence, aberrations, detector response, and sample uncertainty belong to the image forward model.
Quantum-dot and qubit models project full device solutions into a few states. Schrödinger–Poisson or atomistic eigenstates determine orbital, valley, and tunnel couplings; spin and control terms form an effective Hamiltonian. Leakage, charge noise, hyperfine fields, and pulse transfer govern experiments. A two-level Schrödinger evolution is credible only across the calibrated pulse envelope.
Optical transition strengths require wavefunctions as well as energies. Dipole or momentum matrix elements, polarization, occupation, excitons, phonons, and selection rules determine spectra. Envelope overlap controls interband and intersubband response. Broadening and lifetime are open-system properties rather than direct outputs of a closed stationary equation.
**Verification should combine analytic cases, conservation, and systematic refinement.** Recover free-particle dispersion, square-well levels, harmonic-oscillator energies, delta-well matching, and known tunneling limits. Check Hermiticity, norm, current continuity, orthogonality, residuals, gauge consistency, and order of convergence. Refine domain, grid, basis, timestep, absorber, and nonlinear tolerance separately.
Discrete dispersion analysis reveals grid error before device simulation. A second-difference kinetic operator has a cosine dispersion that deviates from $\hbar^2k^2/(2m)$ near the grid Nyquist limit. Requiring several points per shortest wavelength is necessary but observable-specific convergence is stronger. High-energy spurious modes can contaminate driven dynamics even when low states converge.
Domain convergence matters for weakly bound and resonant states. Increase exterior padding and absorber thickness, then compare energies, decay, reflection, and interior observables. A stable eigenvalue in a finite box may track a box mode rather than a resonance. Stabilization methods or complex scaling distinguish them more reliably.
Self-consistent convergence should monitor total charge, Poisson residual, eigenlevel shifts, occupation, current, and free-energy or potential behavior where applicable. Multiple solutions and hysteresis may be physical or numerical. Continuation in bias and multiple initial guesses expose branches. Aggressive mixing can converge to a smoothed but incorrect state.
**Validation must map wavefunctions into actual measured observables.** Compare transition energies and oscillator strengths to spectra, subband occupancy to capacitance or density, transmission to conductance, leakage to current–voltage data, and spatial density to microscopy through instrument response. Absolute wavefunction phase is not measured directly. Calibration and validation datasets should be separated.
Parameter provenance governs prediction. Effective masses, offsets, dielectric constants, strain potentials, interface conditions, disorder distributions, and contact self-energies vary with process, composition, temperature, and orientation. Fitting them all to one curve produces nonunique models. Independent material and geometry measurements reduce compensation.
Uncertainty can be amplified exponentially in tunneling and sharply near avoided crossings. Propagate thickness, barrier height, mass, field, roughness, and temperature distributions rather than only nominal values. Track subspaces when levels reorder. Numerical error and parameter uncertainty should not be merged: refinement reduces one but not the other.
The correct Schrödinger formulation depends on the physical question.
| Question | Equation and representation | Essential extensions | Validation target |
|---|---|---|---|
| Bound level in a well | stationary effective-mass eigenproblem | finite offsets, mass ordering, domain | spectroscopy and mesh convergence |
| Wavepacket motion | time-dependent initial-value problem | absorber, drive, timestep control | norm, current and arrival distribution |
| Barrier transmission | stationary scattering or wavepacket propagation | flux normalization and open boundaries | analytic limit and measured current |
| MOS confinement | Poisson–Schrödinger subband solve | occupations, valleys, fixed charge, temperature | capacitance and charge centroid |
| Coherent device current | open Schrödinger/NEGF problem | leads, self-energies, electrostatic feedback | current and differential conductance |
| Spin or valley control | multicomponent time-dependent equation | noise, leakage, pulse transfer | Rabi, Ramsey, spectroscopy |
| Optical transition | electron–hole or excitonic eigenproblem | dipoles, occupation, phonons, broadening | polarized spectrum and lifetime |
| Many-electron state | interacting configuration-space equation or reduction | antisymmetry and correlation method | energies, densities and correlations |
```flowchart
flowchart TD
A[Define particle model, device, preparation, observable, and tolerance] --> B[Choose Hilbert space, components, Hamiltonian, and operator domain]
B --> C{Stationary spectrum or time evolution?}
C -->|Stationary| D[Specify bound, periodic, or scattering boundary conditions]
C -->|Time evolution| E[Specify normalized initial state, drive, and open boundaries]
D --> F{Is electrostatic or many-body feedback important?}
E --> F
F -->|Yes| G[Couple Poisson, interactions, occupations, or environment self-consistently]
F -->|No| H[Assemble linear Schrödinger problem]
G --> I[Choose grid, basis, finite elements, Green function, or propagator]
H --> I
I --> J[Verify domain, Hermiticity, norm, current, analytic limits, and convergence]
J --> K[Map states through contacts, selection rules, occupations, and instrument]
K --> L[Validate held-out observables with parameter and model uncertainty]
L --> M{Adequate across bias, geometry, temperature, and time?}
M -->|No| N[Revise scale, boundaries, physics, resolution, or parameters]
N --> B
M -->|Yes| O[Deploy with provenance and validity limits]
```
**A reliable solution workflow treats domain and observable as equal to the equation.** Define state preparation and modeled degrees of freedom, build a self-adjoint closed Hamiltonian or declared open extension, impose current-consistent boundaries, and choose stationary or time-dependent numerics. Verify analytic limits and conservation before fitting parameters. Then propagate occupations and instrument response to the measured quantity and validate outside calibration conditions.
```svg
```
Dimensional analysis provides an early error screen. The kinetic term has energy units, wavefunctions carry inverse square-root volume under ordinary normalization, probability current carries probability per area per time, and a delta potential has dimension-dependent coupling units. Nondimensionalization with characteristic length $L$, energy $\hbar^2/(2mL^2)$, and time $\hbar/E$ improves conditioning and reveals controlling ratios. Code should convert back to declared physical units only at interfaces and reports.
Coordinate transformations alter the Laplacian, integration measure, and boundary geometry together. Cylindrical and spherical equations contain metric factors; radial substitutions can remove first derivatives while changing normalization. Curvilinear finite elements encode geometry in Jacobians. Copying a Cartesian kinetic stencil onto a nonuniform or curved coordinate without the correct divergence form breaks self-adjointness and current conservation.
Moving meshes or time-dependent bases add connection terms because basis functions themselves evolve. Expanding $|\psi\rangle=\sum_nc_n(t)|\phi_n(t)\rangle$ produces matrix elements of $i\hbar\langle\phi_m|\dot\phi_n\rangle$ in addition to the projected Hamiltonian. Omitting them creates basis-dependent dynamics. Adiabatic representations, molecular dynamics, and moving quantum dots use these derivative couplings.
Mixed quantum–classical simulation couples Schrödinger amplitudes to classical nuclei, fields, circuits, or mechanics. Ehrenfest dynamics uses mean forces, surface hopping adds stochastic transitions, and Born–Oppenheimer motion selects potential surfaces under separation assumptions. Energy exchange and detailed balance depend on the coupling algorithm. No hybrid method is automatically exact merely because each isolated subsystem uses a standard equation.
Device variability changes both potential and domain. Line-edge roughness, alloy randomness, interface steps, discrete dopants, trapped charge, and thickness variation create ensembles of Schrödinger problems. Averaging potentials before solving is not generally equivalent to averaging observables after solving because eigenvalues and tunneling are nonlinear. Statistical convergence requires enough disorder realizations and a defensible spatial correlation model.
Mesh adaptation should use estimators tied to wavefunction energy, interface flux, or target observables. Refining only where $|\psi|$ is large can miss evanescent regions controlling tunneling. Refining only sharp potentials can waste degrees if the wavefunction is negligible there. Goal-oriented error estimates use an adjoint problem to weight residuals by the measurement of interest.
Parallel solvers partition spatial domains, basis vectors, energy points, bias points, or disorder realizations. Communication boundaries must preserve Hermiticity and flux. Independent energy or sample parallelism is simple; self-consistent Poisson coupling and orthogonalization can dominate synchronization. Performance optimization should retain reproducible convergence tests because altered reduction order changes floating-point results near degeneracy.
Reproducible reporting includes potential zero, coordinate axes, charge sign, mass tensor, basis ordering, boundary conditions, domain size, mesh, timestep, solver and tolerance, normalization, occupations, temperature, broadening, contacts, and extracted observable. It also records whether energies are absolute, relative to a band edge, or referenced to a chemical potential. Without these details, two correct solutions can appear inconsistent or two inconsistent solutions can appear to agree after an arbitrary offset.
Model governance matters when the equation becomes part of a production or design pipeline. Version the material library, geometry source, meshing rules, boundary templates, solver, post-processing, and calibration dataset as one artifact. Regression tests should include analytic benchmarks, representative devices, difficult interfaces, and conservation thresholds. Monitor deployment inputs for extrapolation beyond calibrated bias, temperature, composition, thickness, energy, and field. When a model is updated, compare not only final current or energy but intermediate potential, density, occupation, and wavefunction subspaces so compensating changes do not conceal a broken component. Preserve raw measurements and uncertainty definitions so future parameter updates can be separated from changed preprocessing.
Erwin Schrödinger introduced his wave equation in 1926, building on de Broglie’s matter waves and Hamilton–Jacobi analogies; Max Born supplied the probability interpretation; Werner Heisenberg’s matrix mechanics offered an equivalent formulation; Paul Dirac unified transformation and bra–ket methods; John von Neumann formalized Hilbert-space and operator foundations; Ehrenfest linked expectation dynamics to classical form; WKB carries the names Wentzel, Kramers, and Brillouin; Fermi developed transition-rate theory; Crank and Nicolson supplied a widely used centered time discretization; Hartree and Fock developed self-consistent many-electron approximations; Landauer connected coherent transmission with conductance.
**Schrödinger-equation intuition improves when preparation, current, and boundaries stay visible.** Ask which amplitudes are admissible, how the Hamiltonian and domain generate them, where probability flows, which stationary or transient problem is being solved, what environment or contacts were eliminated, and how the detector converts state into data. Eigenvalues alone are not the prediction. Read the Schrödinger equation through a state-domain-and-probability-flow lens rather than a wave-formula-and-energy-level lens.