Structure Factors

A structure factor is a correlation function expressed in momentum space, with an optional frequency or energy resolution. For a Hermitian lattice operator $O_j$, define

$$O_{\mathbf q} = \sum_{j=1}^{L} e^{-i\mathbf q\cdot\mathbf r_j} O_j.$$

One connected static convention is

$$S_O^c(\mathbf q) = \frac{1}{L} \left\langle \delta O_{\mathbf q} \delta O_{-\mathbf q} \right\rangle,$$

where

$$\delta O_{\mathbf q} = O_{\mathbf q} - \left\langle O_{\mathbf q} \right\rangle.$$

The corresponding dynamic structure factor is

$$S_O^c(\mathbf q,\omega) ={} \frac{1}{2\pi L} \int_{-\infty}^{\infty} dt\, e^{i\omega t} \times \left\langle \delta O_{\mathbf q}(t) \delta O_{-\mathbf q}(0) \right\rangle.$$

With these choices,

$$\int_{-\infty}^{\infty} d\omega\, S_O^c(\mathbf q,\omega) = S_O^c(\mathbf q).$$

The static object records how correlations are distributed over wavelengths. The dynamic object also resolves the energies, lifetimes, and continua of the states reached by the chosen operator. A measured scattering intensity is not automatically equal to either one: probe form factors, polarization projectors, kinematic factors, backgrounds, and instrumental resolution intervene.

Helpful background. Equal-time correlations supplies the operator channels; time-dependent correlations explains the transition frequencies being resolved.

Structure-factor conventions

We use stationary states, $[\rho,H]=0$, and finite second moments. Lattice normalization is per site ($L$ sites); continuum density normalization is per particle ($N$ particles, or $\langle N\rangle$ in a fluctuating-number ensemble). Spins below are dimensionless, $\mathbf s=\mathbf S/\hbar$.

The Fourier phases are $e^{-i\mathbf q\cdot\mathbf r}$ and $e^{+i\omega t}$. The probe transfers $\hbar\mathbf q$ and $\hbar\omega$ to the target, so positive frequency means target excitation. A superscript $c$ means subtraction of one-point products. Delta functions are exact unless a resolution or plotting width is specified.

Other normalizations are legitimate: compare the complete definitions, including the operator channel and whether the variable is energy or angular frequency. The energy-variable Jacobian is derived below.

Static Structure Factor on a Lattice

For Hermitian $O_j$,

$$O_{\mathbf q}^{\dagger} = O_{-\mathbf q}.$$

The full static structure factor is

$$S_O(\mathbf q) = \frac{1}{L} \left\langle O_{\mathbf q} O_{-\mathbf q} \right\rangle.$$

Expanding the modes gives

$$S_O(\mathbf q) ={} \frac{1}{L} \sum_{i,j} e^{-i\mathbf q\cdot (\mathbf r_i-\mathbf r_j)} \times \left\langle O_iO_j \right\rangle.$$

The connected version is

$$S_O^c(\mathbf q) ={} S_O(\mathbf q) - \frac{1}{L} \left\langle O_{\mathbf q} \right\rangle \left\langle O_{-\mathbf q} \right\rangle.$$

For Hermitian local operators, the subtracted term is

$$\frac{1}{L} \left| \left\langle O_{\mathbf q} \right\rangle \right|^2.$$

This term can be macroscopically large at an ordering wavevector. Removing it is appropriate when the question concerns fluctuations about a selected profile; retaining it is appropriate when the question concerns elastic order.

Positivity

For every $\mathbf q$,

$$S_O(\mathbf q) = \frac{1}{L} \left\langle O_{\mathbf q} O_{\mathbf q}^{\dagger} \right\rangle \geq 0.$$

The connected version is also nonnegative:

$$S_O^c(\mathbf q) = \frac{1}{L} \left\langle \delta O_{\mathbf q} \delta O_{\mathbf q}^{\dagger} \right\rangle \geq 0.$$

A negative scalar static structure factor signals a sign, ordering, normalization, subtraction, or numerical error. Cross-component tensor entries need not be individually positive; the complete tensor is constrained instead.

Translation invariance

If

$$C_O^c(\mathbf r) = \left\langle \delta O_{j+\mathbf r} \delta O_j \right\rangle$$

depends only on the displacement, then

$$S_O^c(\mathbf q) = \sum_{\mathbf r} e^{-i\mathbf q\cdot\mathbf r} C_O^c(\mathbf r).$$

For a periodic finite lattice, the inverse transform is

$$C_O^c(\mathbf r) = \frac{1}{L} \sum_{\mathbf q} e^{i\mathbf q\cdot\mathbf r} S_O^c(\mathbf q),$$

where the sum is over allowed reciprocal-lattice momenta. At zero displacement,

$$\frac{1}{L} \sum_{\mathbf q} S_O^c(\mathbf q) = C_O^c(\mathbf 0).$$

This onsite sum is a strong numerical normalization check.

Open boundaries, traps, domains, and disorder break ordinary translation invariance. A displacement average can still be Fourier transformed, but it discards center-position information. The subtraction of local means must occur before that average.

Tensor and Channel Structure Factors

For spin components,

$$S^{\alpha\beta}(\mathbf q) ={} \frac{1}{L} \sum_{i,j} e^{-i\mathbf q\cdot (\mathbf r_i-\mathbf r_j)} \times \left\langle \delta s_i^\alpha \delta s_j^\beta \right\rangle.$$

At fixed $\mathbf q$, the component matrix obeys

$$\left[ S^{\alpha\beta}(\mathbf q) \right]^* = S^{\beta\alpha}(\mathbf q)$$

under the stated ordering. For every complex vector $v_\alpha$,

$$\sum_{\alpha,\beta} v_\alpha^* S^{\alpha\beta}(\mathbf q) v_\beta \geq 0.$$

Thus the tensor is positive semidefinite even though an off-diagonal entry can be complex or negative.

Different channels answer different physical questions:

Operator	Typical structure factor	Information emphasized
$n_j$	charge or density	compressibility, density waves, sound, particle–hole continua
$s_j^\alpha$	spin tensor	magnetic order, magnons, spinons, anisotropy
$n_{j\uparrow}-n_{j\downarrow}$	longitudinal spin	magnetic density in fermion models
$\Delta_j$	pair	pair coherence and finite-momentum pairing
bond energy $B_j$	bond or dimer	valence-bond order and singlet fluctuations
current $j_j^\alpha$	current spectrum	transport and continuity constraints

A probe couples to a particular operator and often to only a tensor projection. Peaks in two channels at the same momentum do not imply identical matrix elements or dynamics.

Charge and Spin Density Waves explains how these channels help distinguish finite-wavevector order from short-range fluctuations in materials.

Continuum Density Structure Factor

For $N$ point particles,

$$\rho_{\mathbf q} = \sum_{j=1}^{N} e^{-i\mathbf q\cdot\mathbf r_j}.$$

A connected per-particle convention is

$$S_n^c(\mathbf q) = \frac{1}{N} \left\langle \delta\rho_{\mathbf q} \delta\rho_{-\mathbf q} \right\rangle.$$

For a homogeneous single-species fluid with positive mean density, define the normal-ordered pair distribution

$$g^{(2)}(\mathbf r) = \frac{ \left\langle : \hat n(\mathbf r) \hat n(\mathbf 0) : \right\rangle }{ n^2 },$$

where $n=N/V$. In the thermodynamic limit,

$$S_n^c(\mathbf q) ={} 1 + n \int d^d r\, e^{-i\mathbf q\cdot\mathbf r} \left[ g^{(2)}(\mathbf r)-1 \right].$$

The formula also holds in a finite periodic box when the finite-box pair distribution and integral are used consistently. Taking the thermodynamic limit inside the integral at exactly zero momentum can discard the small, spatially extended contribution imposed by fixed particle number. The leading $1$ is the self term $i=j$. The integral contains distinct-particle correlations. Normal ordering in $g^{(2)}$ prevents the self delta function from being counted twice.

For short-range correlations, the point-particle structure factor often approaches $1$ at large $|\mathbf q|$. A measured intensity need not do so because atomic, molecular, magnetic, or Wannier form factors can decay with momentum.

The zero-momentum subtlety

Because

$$\rho_{\mathbf 0} = N,$$

the full structure factor at zero momentum is

$$S_n(\mathbf 0) = \frac{\langle N^2\rangle}{\langle N\rangle}$$

when particle number can fluctuate. The connected value is

$$S_n^c(\mathbf 0) = \frac{ \operatorname{Var}(N) }{ \langle N\rangle }.$$

In a fixed-$N$ ensemble it vanishes exactly. The thermodynamic limit $\mathbf q\to\mathbf 0$ through nonzero wavevectors can differ from evaluating the exactly conserved mode first. This is an order-of-limits and ensemble issue, not a contradiction.

Dynamic Structure Factor

For a stationary state,

$$S_O(\mathbf q,\omega) ={} \frac{1}{2\pi L} \int_{-\infty}^{\infty} dt\, e^{i\omega t} \times \left\langle O_{\mathbf q}(t) O_{-\mathbf q}(0) \right\rangle.$$

The connected definition replaces each mode by its centered version. The quantity is an ordinary ordered correlation spectrum. It is not a retarded response function and contains no step function or commutator.

For spin components, the corresponding tensor is

$$S^{\alpha\beta}(\mathbf q,\omega) ={} \frac{1}{2\pi L} \int_{-\infty}^{\infty} dt\, e^{i\omega t} \times \left\langle s_{\mathbf q}^{\alpha}(t) s_{-\mathbf q}^{\beta}(0) \right\rangle.$$

At each $\omega$, this tensor is Hermitian and positive semidefinite as a distribution for conjugate operator channels. Experimental polarization factors select particular quadratic forms of it.

Frequency versus energy transfer

The convention above uses angular frequency. If an experiment reports the target energy transfer

$$E = \hbar\omega$$

and the energy-resolved structure factor is normalized by

$$\int dE\, S_O(\mathbf q,E) = S_O(\mathbf q),$$

then

$$S_O(\mathbf q,E) = \frac{1}{\hbar} S_O\!\left( \mathbf q, \frac{E}{\hbar} \right).$$

Consequently, a definition with $1/(2\pi\hbar)$ in the time transform is usually energy-resolved, while one with $1/(2\pi)$ is usually angular-frequency-resolved. Units expose a missing Jacobian.

Lehmann Representation

Let

$$\rho = \sum_n p_n \lvert n\rangle\langle n\rvert, \qquad H\lvert n\rangle = E_n\lvert n\rangle.$$

Insertion of a complete energy basis gives

$$S_O(\mathbf q,\omega) ={} \frac{1}{L} \sum_{n,m} p_n \left| \left\langle m \middle| O_{-\mathbf q} \middle| n \right\rangle \right|^2 \times \delta\!\left( \omega- \frac{E_m-E_n}{\hbar} \right).$$

Every line has:

• an initial-state weight $p_n$;
• a nonnegative operator matrix element;
• momentum transfer fixed by $O_{-\mathbf q}$;
• a transition frequency $(E_m-E_n)/\hbar$.

For the conjugate operator pair, $S_O(\mathbf q,\omega)$ is a nonnegative distribution. At zero temperature, after removing any elastic ground-state contribution,

$$S_O(\mathbf q,\omega) = 0, \qquad \omega<0,$$

because the ground state cannot release energy it does not possess.

At finite temperature, negative-frequency weight describes target de-excitation: the probe can gain energy from an initially excited target.

Zeroth Moment and Equal Time

Integrating the Lehmann representation gives

$$\begin{aligned} \int d\omega\, S_O(\mathbf q,\omega) &= \frac{1}{L} \left\langle O_{\mathbf q} O_{-\mathbf q} \right\rangle \\ &= S_O(\mathbf q). \end{aligned}$$

This zeroth-moment identity is exact under the declared normalization. A numerical spectrum that fails it has lost weight through finite frequency range, broadening normalization, time-windowing, analytic continuation, or inconsistent static and dynamic conventions.

Energy-integrated experimental data equal the static structure factor only after all relevant energy transfers are captured and the same probe factors are removed. A finite detector window produces a partial moment.

Detailed Balance

For a thermal state,

$$p_n = \frac{e^{-\beta E_n}}{Z}.$$

Swapping the initial and final states in the Lehmann sum gives

$$S_O(-\mathbf q,-\omega) = e^{-\beta\hbar\omega} S_O(\mathbf q,\omega).$$

If inversion symmetry also implies

$$S_O(-\mathbf q,\omega) = S_O(\mathbf q,\omega),$$

then

$$S_O(\mathbf q,-\omega) = e^{-\beta\hbar\omega} S_O(\mathbf q,\omega).$$

The energy-gain side is thermally suppressed relative to the energy-loss side. This is both a physical statement and a calibration check, but it applies only after background subtraction and with a consistent sign convention for target energy transfer.

Elastic, Quasielastic, and Inelastic Weight

For a stationary one-point function,

$$\left\langle O_{\mathbf q}(t) \right\rangle = \left\langle O_{\mathbf q} \right\rangle.$$

Therefore the full spectrum separates as

$$S_O(\mathbf q,\omega) ={} \frac{ \left| \left\langle O_{\mathbf q} \right\rangle \right|^2 }{L} \delta(\omega) + S_O^c(\mathbf q,\omega).$$

The first term is an exactly elastic coherent contribution. In a crystal or selected ordered phase it can produce a Bragg peak.

Connected subtraction removes that one-point product, but it need not remove every zero-frequency contribution. Degenerate transitions, conserved projections, phase mixtures, frozen disorder, and finite-volume symmetry sectors can leave connected weight at $\omega=0$.

Terminology:

Spectral feature	Ideal mathematical form	Typical interpretation
elastic	$\delta(\omega)$	static order or time-independent profile
quasielastic	narrow finite-width peak around $0$	slow relaxation, diffusion, domains, or unresolved dynamics
inelastic mode	peak at $\omega\neq0$	excitation with a characteristic energy
continuum	weight over a finite region	multi-particle states, decay, fractionalization, or disorder

A resolution-limited peak is not proof of an exact delta function. It means only that any intrinsic width lies below the instrument’s resolving power.

Momentum–Energy Map

The probe transfers $\hbar\mathbf q=\hbar(\mathbf k_i-\mathbf k_f)$ and $\hbar\omega=E_i-E_f$ to the target. In the resulting $S(\mathbf q,\omega)$ map, elastic order lies at $\omega=0$, a stable collective excitation traces a narrow dispersion, and multi-particle states occupy a continuum. Actual data are multiplied by probe factors and convolved with instrumental resolution.

The two axes resolve different structures:

• a peak position in $\mathbf q$ identifies a dominant wavelength;
• a momentum width of order $\kappa$ often corresponds to a real-space length of order $\kappa^{-1}$;
• a dispersing peak traces an excitation energy $\hbar\omega_{\mathbf q}$;
• an intrinsic frequency width of order $\Gamma$ can correspond to a lifetime of order $\Gamma^{-1}$;
• a continuum boundary records multi-particle kinematics or selection rules;
• redistribution of weight can matter even when peak positions barely move.

The inverse-width statements are scaling estimates, not universal equalities. Line shapes, dimensionality, algebraic prefactors, disorder, and resolution determine the precise conversion.

From Correlations to a Scattering Cross Section

Suppose an incident probe has wavevector $\mathbf k_i$ and energy $E_i$, while the detected probe has $\mathbf k_f$ and $E_f$. This page defines target transfers by

$$\mathbf q = \mathbf k_i-\mathbf k_f, \qquad \hbar\omega = E_i-E_f.$$

Thus $\omega>0$ means the target absorbs energy.

In the first Born or golden-rule regime, a probe that couples linearly to a target operator produces the schematic factorization

$$\frac{d^2\sigma}{ d\Omega\,dE_f } = \mathcal K \, \left| F(\mathbf q) \right|^2 \, \mathcal P \!\left[ S_O(\mathbf q,\omega) \right].$$

Here:

• $\mathcal K$ contains flux and kinematic factors such as $k_f/k_i$;
• $F(\mathbf q)$ is a single-particle, orbital, magnetic, molecular, or Wannier form factor;
• $\mathcal P$ is a scalar or tensor projection set by probe polarization and coupling;
• $S_O$ contains the target’s many-body correlations.

This factorization is not universal beyond weak single-scattering conditions. Multiple scattering, strong final-state interactions, absorption, refraction, detector nonlinearity, and probe-specific vertices can require a more complete forward model.

Neutron Scattering

Neutrons couple both to nuclei and to magnetic moments. Those channels expose different target operators.

Nuclear scattering

Coherent nuclear scattering retains interference among different nuclei and therefore carries collective spatial structure. In the simplest identical-scatterer limit,

$$\frac{d^2\sigma_{\mathrm{coh}}}{ d\Omega\,dE_f } \propto \frac{k_f}{k_i} S_n(\mathbf q,E_i-E_f).$$

Real materials contain isotope- and nuclear-spin-dependent scattering lengths. Their variance produces an incoherent contribution dominated by self correlations. Incoherent scattering is not merely a featureless nuisance: it can measure single-particle motion, but it is not the same collective pair structure factor as the coherent term.

Magnetic scattering

For unpolarized magnetic neutron scattering, a standard schematic form is

$$\frac{d^2\sigma_{\mathrm{mag}}}{ d\Omega\,dE_f } \propto{} \frac{k_f}{k_i} \left| F_{\mathrm{mag}}(\mathbf q) \right|^2 \times \sum_{\alpha,\beta} \left( \delta_{\alpha\beta} - \widehat q_\alpha \widehat q_\beta \right) S^{\alpha\beta}(\mathbf q,\omega).$$

The projector removes magnetization parallel to $\mathbf q$: the neutron is sensitive to components transverse to the momentum transfer. The magnetic form factor reflects the spatial distribution of electronic magnetization and usually decreases at large momentum.

Polarized-neutron analysis can separate additional symmetric and antisymmetric tensor combinations. Crystal domains, multiple magnetic ions per cell, $g$-tensor anisotropy, and nuclear–magnetic interference require corresponding structure and polarization factors.

Neutron Scattering explains beam kinematics, instrument resolution, isotope and polarization choices, and measurements of phonons, magnons, and continua.

Cold-Atom Probes

Cold-atom platforms realize both frequency-resolved and equal-time measurements without a material crystal.

Two-photon Bragg spectroscopy

Two laser beams with wavevectors $\mathbf k_1,\mathbf k_2$ and frequencies $\omega_1,\omega_2$ create a moving optical potential. The target transfers are

$$\mathbf q = \mathbf k_1-\mathbf k_2, \qquad \omega = \omega_1-\omega_2.$$

In weak driving, the absorption rate at positive $\omega$ is proportional to a density or spin dynamic structure factor, depending on the light coupling. A measured net momentum or energy transfer generally involves the difference between absorption and stimulated emission:

$$S(\mathbf q,\omega) - S(-\mathbf q,-\omega).$$

At low temperature and positive target energy transfer, the reverse term can be strongly suppressed. At finite temperature it must not be silently discarded.

Finite pulse duration broadens frequency resolution. Trap inhomogeneity averages regions with different density and gap scales. Strong pulses cause saturation and nonlinear response. Final-state interactions and optical selection rules can change the operator actually measured.

In particular, a finite-pulse momentum-transfer spectrum in a trapped condensate is a probe response built from the dynamic structure factor, not generally the dynamic structure factor itself. The pulse envelope and inhomogeneous evolution belong in the forward model.

Density snapshots and noise correlations

Quantum gas microscopes and absorption images can estimate equal-time density correlations. Fourier transformation of connected fluctuations produces a static structure factor:

$$\widehat S_n^c(\mathbf q) \propto \left\langle \left| \delta n_{\mathbf q} \right|^2 \right\rangle.$$

Finite imaging resolution multiplies the mode by an optical transfer function. Atom loss, parity projection, finite field of view, trap profiles, and shot-to-shot number fluctuations must be modeled before comparison with a homogeneous theory.

X-Ray and Electron Probes

Elastic and inelastic x-ray scattering primarily couple to charge density, modulated by atomic form factors and photon polarization. Diffraction emphasizes static periodic density, while inelastic x-ray scattering resolves charge dynamics over momentum and energy.

X-Ray Scattering develops the photon kinematics, elastic and resonant amplitudes, absorption corrections, coherent phase retrieval, charge-order workflow, and experimental distinction between nonresonant IXS and RIXS.

Electron energy-loss spectroscopy often reports a loss function such as

$$- \operatorname{Im} \frac{1}{ \epsilon(\mathbf q,\omega) },$$

not a bare structure factor. The loss function, density response, and $S_n(\mathbf q,\omega)$ are related through electrodynamics and fluctuation–dissipation relations under stated conditions, but they are not interchangeable labels.

Plasmons Preview applies this distinction to collective charge modes, including dielectric zeros, loss peaks, and density pole weights.

Raman scattering, resonant x-ray scattering, and optical spectroscopy can couple to stress, spin, orbital, bond, or symmetry-resolved composite operators. “Scattering measures $S$” is only meaningful after the operator and vertex are specified.

Order, Criticality, and Peak Scaling

The full phase criterion and its finite-size caveats are developed in Long-Range Order. This section records how those regimes appear in the present Fourier convention.

Let $O_j$ have a commensurate ordering wavevector $\mathbf Q$ with real phase $e^{i\mathbf Q\cdot(\mathbf r_i-\mathbf r_j)}=\pm1$. If a full equal-time correlator approaches a nonzero plateau,

$$\left\langle O_iO_j \right\rangle \longrightarrow m^2 e^{i\mathbf Q\cdot (\mathbf r_i-\mathbf r_j)},$$

then

$$S_O(\mathbf Q) \sim L m^2.$$

The peak height is extensive under the $1/L$ convention. Equivalently,

$$\frac{ S_O(\mathbf Q) }{L} \longrightarrow m^2$$

after the correct thermodynamic and phase-selection limits. For incommensurate real order, resolve both $+\mathbf Q$ and $-\mathbf Q$ harmonics; a cosine can split the weight between them, so the coefficient must be derived from the specified profile.

Short-range connected correlations produce a peak whose height is controlled by a correlation volume rather than the whole system. Landau–Ginzburg Theory Preview derives the Gaussian Lorentzian peak and its width $\xi^{-1}$; the normalization and experimental qualifications remain here. Near a critical point, if

$$C_O^c(r) \sim r^{-2\Delta_O},$$

then a finite system of linear size $\ell$ can show

$$S_O^c(\mathbf Q) \sim \ell^{d-2\Delta_O}$$

when the infrared integral diverges with size. Marginal cases can carry logarithms.

Critical Exponents and Scaling relates this peak power to $\eta$, $\nu$, and $\gamma$, including finite-size corrections.

Peak height alone is insufficient. Reliable phase inference combines:

• height scaling with $L$ or $\ell$;
• peak-width scaling and a correlation-length estimator;
• the full versus connected convention;
• boundary and aspect-ratio dependence;
• tensor and domain projections;
• an order-parameter or Binder-style cross-check.

Example: Uncorrelated Uniform Particles

For independent distinguishable particles, or a classical ideal gas, uniformly distributed in a periodic volume and at a nonzero allowed momentum,

$$\left\langle e^{-i\mathbf q\cdot\mathbf r_j} \right\rangle = 0.$$

Then

$$\begin{aligned} S_n(\mathbf q) &= \frac{1}{N} \sum_{i,j} \left\langle e^{-i\mathbf q\cdot (\mathbf r_i-\mathbf r_j)} \right\rangle \\ &= \frac{1}{N} \sum_i 1 \\ &= 1. \end{aligned}$$

The result is the self term. Bose or Fermi exchange, interactions, a trap, or a global constraint changes the cross terms even when no classical pair force is present.

At $\mathbf q=\mathbf 0$, the argument fails because every phase equals one. The full fixed-$N$ value is $N$, whereas the connected value is zero.

Example: Néel Order and Phase Selection

Consider an even one-dimensional chain with

$$\left\langle s_j^z \right\rangle_+ = m(-1)^j, \qquad \mathbf Q = \frac{\pi}{a}.$$

Take each site to be an eigenstate of $s_j^z$, with eigenvalue $m(-1)^j$ (for fully polarized spin-$1/2$ sites, $m=1/2$). In this selected product pattern,

$$\left\langle s_{\mathbf Q}^z \right\rangle_+ = Lm.$$

Therefore

$$S_{\mathrm{full}}^{zz}(\mathbf Q) = Lm^2,$$

while the connected structure factor of the exact product pattern vanishes because there are no fluctuations about its local means.

An equal mixture of the two translated Néel patterns has

$$\left\langle s_j^z \right\rangle_{\mathrm{mix}} = 0$$

but retains

$$S_{\mathrm{mix}}^{zz}(\mathbf Q) = S_{\mathrm{mix},c}^{zz}(\mathbf Q) = Lm^2.$$

The same sharp static peak can therefore coexist with different one-point functions and different clustering properties. Structure-factor data alone do not determine whether the state is a selected phase, a symmetric mixture, or a coherent finite-volume state.

Example: Antiferromagnetic Spin Dimer

For two spin-$1/2$ sites separated by $a$ in a singlet,

$$\left\langle \mathbf s_1^2 \right\rangle = \left\langle \mathbf s_2^2 \right\rangle = \frac{3}{4},$$

and

$$\left\langle \mathbf s_1\cdot\mathbf s_2 \right\rangle = - \frac{3}{4}.$$

With $L=2$,

$$\begin{aligned} S(\mathbf q) &= \frac{1}{2} \sum_{i,j} e^{-i\mathbf q\cdot (\mathbf r_i-\mathbf r_j)} \left\langle \mathbf s_i\cdot\mathbf s_j \right\rangle \\ &= \frac{3}{4} \left[ 1-\cos(qa) \right]. \end{aligned}$$

Thus

$$S(0) = 0, \qquad S(\pi/a) = \frac{3}{2}.$$

The zero-momentum cancellation expresses total-spin singlet character. The antiferromagnetic momentum maximizes the contrast between opposite spins. For the Heisenberg dimer, the dynamic spin structure factor places the corresponding inelastic weight at the singlet–triplet gap.

Modes, Continua, and Spectral Weight

A sharp quasiparticle contribution has the form

$$S_O(\mathbf q,\omega) \supset Z_O(\mathbf q) \delta\!\left( \omega-\omega_{\mathbf q} \right).$$

The residue

$$Z_O(\mathbf q)$$

depends on the operator. A real excitation can be absent from a chosen structure factor because symmetry makes its matrix element vanish.

Interactions can broaden a mode or move weight into a continuum:

$$S_O = S_O^{\mathrm{mode}} + S_O^{\mathrm{cont}}.$$

The integrated weight must still satisfy exact moments. Calling a broad feature “a particle” without checking size, resolution, thresholds, and sum rules can misidentify its physical origin. Phonons as Many-Body Excitations derives the displacement matrix element, polarization factor, and creation–annihilation weights behind the harmonic one-phonon peak.

Particle–Hole Excitations derives the occupied-to-empty promotion, Pauli-constrained continuum boundaries, and the independent density response whose absorptive part supplies a canonical fermionic continuum.

Bogoliubov Quasiparticles shows how particle–hole amplitudes combine into density, spin, current, and pair-channel coherence factors before they become the operator-dependent weight $Z_O$.

The Heisenberg Model provides a standard contrast: higher-dimensional ordered magnets can show sharp magnons, while a one-dimensional spin-$1/2$ antiferromagnet has prominent spinon continua. Magnons derives the transverse one-magnon matrix elements, sublattice coherence factors, and longitudinal multi-magnon onset. Goldstone Modes in Many-Body Systems explains when a soft pole is enforced by broken symmetry and how its finite-size and low-momentum scaling should be tested.

First Moment and the Density f-Sum

The first angular-frequency moment is

$$m_1(\mathbf q) = \int d\omega\, \omega S_O(\mathbf q,\omega).$$

For the density channel of a homogeneous continuum system in equilibrium at rest, assume the first moments agree at $\mathbf q$ and $-\mathbf q$ (inversion symmetry is sufficient). Then

$$m_{1,n}(\mathbf q) ={} \frac{1}{ 2\hbar N } \times \left\langle \left[ \rho_{\mathbf q}, \left[ H, \rho_{-\mathbf q} \right] \right] \right\rangle.$$

Without that symmetry the double commutator gives $[m_1(\mathbf q)+m_1(-\mathbf q)]/2$, rather than either moment separately. For nonrelativistic particles of mass $m$ in the stated rest equilibrium, with kinetic energy $\sum_j\mathbf p_j^2/(2m)$ and position-dependent interactions,

$$\int d\omega\, \omega S_n(\mathbf q,\omega) = \frac{ \hbar q^2 }{ 2m }.$$

This is the per-particle density $f$-sum rule in the present angular-frequency convention. Lattices, unequal masses, momentum-dependent interactions, and projected bands require evaluation of their own commutators; the continuum expression should not be assumed. A current-carrying state also shifts the unsymmetrized first moment.

At zero temperature, if one sharp mode exhausts the density weight,

$$S_n(\mathbf q,\omega) = S_n(\mathbf q) \delta\!\left( \omega-\omega_{\mathbf q} \right),$$

then

$$\hbar\omega_{\mathbf q} = \frac{ \hbar^2q^2 }{ 2mS_n(\mathbf q) }.$$

This is the single-mode or Feynman relation under a strong assumption. With a continuum, the ratio $m_1/m_0$ is a weighted mean frequency rather than the energy of every excitation.

Sum Rules derives the moments systematically; here they help check normalization and interpret spectral weight.

Compressibility and Response Boundary

In a homogeneous equilibrium fluid at nonzero temperature, an appropriate thermodynamic limit gives

$$\lim_{\mathbf q\to\mathbf 0} S_n^c(\mathbf q) = n k_{\mathrm B}T \kappa_T,$$

where $\kappa_T$ is the isothermal compressibility. The exactly conserved $\mathbf q=0$ mode in a canonical finite system can still vanish.

The dynamic structure factor is related to the dissipative density response by a fluctuation–dissipation theorem. That relation includes thermal factors and convention-dependent signs; it does not make $S$ itself causal. Kubo Formula derives retarded response, while Fluctuations and Susceptibilities explains thermodynamic fluctuation scaling.

Numerical Evaluation

Equal-time configurations

For periodic lattice samples,

$$S_O^c(\mathbf q) = \frac{1}{L} \left\langle \left| \delta O_{\mathbf q} \right|^2 \right\rangle$$

can be accumulated with a discrete Fourier transform. Subtract local profiles before averaging in inhomogeneous systems. Check positivity, the onsite momentum sum, symmetry-related momenta, and exact conserved modes.

Exact diagonalization

The Lehmann sum produces exact finite-size delta lines. Artificial broadening is useful for plotting but must preserve integrated weight. A chosen Lorentzian width is not a lifetime.

Real-time tensor networks

Compute

$$C_O(\mathbf q,t) = \frac{1}{L} \left\langle O_{\mathbf q}(t) O_{-\mathbf q}(0) \right\rangle$$

and Fourier transform a finite time interval. Entanglement growth limits $t_{\max}$; the resulting frequency resolution is of order $1/t_{\max}$. Window functions trade ringing against broadening and can alter peak height.

Imaginary-time methods

For

$$0\leq\tau\leq\beta\hbar,$$

the imaginary-time density correlator has the spectral form

$$F(\mathbf q,\tau) = \int d\omega\, e^{-\omega\tau} S_n(\mathbf q,\omega).$$

Recovering real-frequency structure from noisy imaginary-time data is an ill-conditioned inverse problem. Maximum entropy, stochastic continuation, sparse models, or neural methods introduce assumptions; agreement with moments and synthetic benchmarks is essential.

Correction-vector and Krylov methods

Frequency-domain resolvents can target selected $\omega$ directly, while Krylov propagation resolves finite-time dynamics. Linear-system tolerance, boundary reflections, bond dimension, and broadening all affect the inferred line shape.

Experimental Forward Model

Measured counts are better represented schematically by

$$I_{\mathrm{meas}}(\mathbf q,\omega) ={} \int d^d q'\,d\omega'\, R( \mathbf q-\mathbf q', \omega-\omega' ) \times M(\mathbf q',\omega') + B(\mathbf q,\omega),$$

where:

• $M$ contains the cross section, structure factor, form factors, and polarization terms;
• $R$ is the multidimensional instrumental resolution;
• $B$ is background.

Deconvolution can be unstable. It is often more reliable to convolve a theoretical model with $R$ and fit it in measurement space.

The relevant corrections depend on the instrument: flux and detector calibration set the scale, while absorption, background, polarization, domain populations, and finite acceptance can reshape the signal. Propagate correlated uncertainties when fitting. How Quantum Matter Is Measured develops these measurement issues in more detail.

Common pitfalls

A connected spectrum can still have elastic weight. Subtracting $\langle O_{\mathbf q}\rangle\langle O_{-\mathbf q}\rangle$ removes a static mean profile. Conserved fluctuations, degeneracies, or a phase mixture can leave a zero-frequency delta function.

A plotting width is not a lifetime. Exact finite-volume lines remain delta functions. Compare several plotting widths and instrumental resolutions before assigning an intrinsic relaxation rate.

A detector window need not contain the full static weight. The equal-time sum rule integrates over all energy transfers. Missing high-energy weight and unremoved probe factors can change the apparent normalization.

Exercises

Exercise 1: Positivity and the momentum sum

Show that

$$S_O^c(\mathbf q) = \frac{1}{L} \left\langle \delta O_{\mathbf q} \delta O_{\mathbf q}^{\dagger} \right\rangle \geq 0.$$

For a translation-invariant periodic lattice, also prove

$$\frac{1}{L} \sum_{\mathbf q} S_O^c(\mathbf q) = \left\langle (\delta O_j)^2 \right\rangle.$$

Solution

For any vector $\lvert\psi\rangle$,

$$\left\langle \psi \middle| \delta O_{\mathbf q} \delta O_{\mathbf q}^{\dagger} \middle| \psi \right\rangle = \left\| \delta O_{\mathbf q}^{\dagger} \lvert\psi\rangle \right\|^2 \geq 0.$$

A density operator is a positive mixture of such vectors, so its expectation remains nonnegative.

Translation invariance gives

$$S_O^c(\mathbf q) = \sum_{\mathbf r} e^{-i\mathbf q\cdot\mathbf r} C_O^c(\mathbf r).$$

Therefore

$$\begin{aligned} \frac{1}{L} \sum_{\mathbf q} S_O^c(\mathbf q) &= \sum_{\mathbf r} C_O^c(\mathbf r) \left[ \frac{1}{L} \sum_{\mathbf q} e^{-i\mathbf q\cdot\mathbf r} \right] \\ &= C_O^c(\mathbf 0) \\ &= \left\langle (\delta O_j)^2 \right\rangle. \end{aligned}$$

The bracket is the discrete reciprocal-lattice delta function.

Exercise 2: Zeroth moment

Starting from

$$S_O(\mathbf q,\omega) ={} \frac{1}{L} \sum_{n,m} p_n \left| \langle m| O_{-\mathbf q} |n\rangle \right|^2 \times \delta\!\left( \omega- \frac{E_m-E_n}{\hbar} \right),$$

show that its integral equals $S_O(\mathbf q)$.

Solution

Integrating the delta function gives

$$\int d\omega\, S_O(\mathbf q,\omega) = \frac{1}{L} \sum_{n,m} p_n \left| \langle m| O_{-\mathbf q} |n\rangle \right|^2.$$

Use completeness:

$$\begin{aligned} \sum_m \left| \langle m| O_{-\mathbf q} |n\rangle \right|^2 &= \langle n| O_{\mathbf q} O_{-\mathbf q} |n\rangle. \end{aligned}$$

Then

$$\int d\omega\, S_O(\mathbf q,\omega) = \frac{1}{L} \operatorname{Tr} \left( \rho O_{\mathbf q} O_{-\mathbf q} \right) = S_O(\mathbf q).$$

Exercise 3: Detailed balance

For a thermal state, derive

$$S_O(-\mathbf q,-\omega) = e^{-\beta\hbar\omega} S_O(\mathbf q,\omega).$$

Solution

Write

$$S_O(-\mathbf q,-\omega) ={} \frac{1}{L} \sum_{n,m} p_n \left| \langle m| O_{\mathbf q} |n\rangle \right|^2 \times \delta\!\left( -\omega- \frac{E_m-E_n}{\hbar} \right).$$

Exchange $n$ and $m$. On the delta-function support,

$$E_m-E_n = \hbar\omega.$$

The Boltzmann weights obey

$$p_m = p_n e^{-\beta(E_m-E_n)} = p_n e^{-\beta\hbar\omega}.$$

The exchanged matrix element is the one appearing in $S_O(\mathbf q,\omega)$, so the stated factor follows.

Exercise 4: Independent-particle baseline

For $N$ independent particles uniformly distributed in a periodic box, show that

$$S_n(\mathbf q) = 1$$

for every nonzero allowed momentum. Why does the same argument fail at $\mathbf q=0$?

Solution

Expand

$$S_n(\mathbf q) = \frac{1}{N} \sum_{i,j} \left\langle e^{-i\mathbf q\cdot\mathbf r_i} e^{i\mathbf q\cdot\mathbf r_j} \right\rangle.$$

For $i=j$, every term is one. For $i\neq j$, independence factorizes the average:

$$\left\langle e^{-i\mathbf q\cdot\mathbf r_i} \right\rangle \left\langle e^{i\mathbf q\cdot\mathbf r_j} \right\rangle.$$

Uniformity makes either factor zero at a nonzero allowed reciprocal momentum. Thus only the $N$ self terms survive and $S_n=1$.

At $\mathbf q=0$, every phase is one, so all $N^2$ terms survive. The full fixed-number value is $N$; the connected value is zero because $\delta N=0$.

Exercise 5: Selected and mixed Néel states

Let each $s_j^z$ have the definite value $m(-1)^j$ in a product state on an even chain. Compute the full and connected structure factors at $Q=\pi/a$ in:

1. one selected product pattern;
2. an equal mixture of the two translated patterns.

Solution

In either selected pattern,

$$s_Q^z = \sum_j e^{-i\pi j} s_j^z = \pm Lm.$$

Hence

$$S_{\mathrm{full}}^{zz}(Q) = \frac{1}{L} \left| \langle s_Q^z\rangle \right|^2 = Lm^2.$$

The local variables have no fluctuations in the product pattern, so

$$S_c^{zz}(Q) = 0.$$

In the equal mixture,

$$\langle s_j^z\rangle_{\mathrm{mix}} = 0,$$

but both components have

$$\left| s_Q^z \right|^2 = L^2m^2.$$

Therefore

$$S_{\mathrm{full}}^{zz}(Q) = S_c^{zz}(Q) = Lm^2.$$

The mixture retains the ordered global label and fails cluster decomposition.

Exercise 6: Spin-dimer structure factor

For a spin-$1/2$ singlet on sites at $0$ and $a$, derive

$$S(q) = \frac{3}{4} \left[ 1-\cos(qa) \right].$$

Check $q=0$ and $q=\pi/a$.

Solution

The onsite terms contribute

$$\frac{1}{2} \left( \frac{3}{4} + \frac{3}{4} \right) = \frac{3}{4}.$$

The two cross terms contribute

$$\frac{1}{2} \left[ e^{iqa} \left( -\frac{3}{4} \right) + e^{-iqa} \left( -\frac{3}{4} \right) \right] = - \frac{3}{4} \cos(qa).$$

Adding onsite and cross terms gives

$$S(q) = \frac{3}{4} \left[ 1-\cos(qa) \right].$$

Thus

$$S(0) = 0,$$

as required for a total-spin singlet, while

$$S(\pi/a) = \frac{3}{2}.$$

Exercise 7: Frequency and energy conventions

Suppose $S_\omega(\mathbf q,\omega)$ satisfies

$$\int d\omega\, S_\omega(\mathbf q,\omega) = S(\mathbf q).$$

Find $S_E(\mathbf q,E)$ such that

$$\int dE\, S_E(\mathbf q,E) = S(\mathbf q),$$

with $E=\hbar\omega$.

Solution

Because

$$dE = \hbar\,d\omega,$$

the two integrands must satisfy

$$S_E(\mathbf q,E)\,dE = S_\omega(\mathbf q,\omega)\,d\omega.$$

Therefore

$$S_E(\mathbf q,E) = \frac{1}{\hbar} S_\omega\!\left( \mathbf q, \frac{E}{\hbar} \right).$$

The factor $1/\hbar$ is required both by normalization and by units.

Exercise 8: Single-mode estimate

At zero temperature, assume the density spectrum is exhausted by

$$S_n(\mathbf q,\omega) = S_n(\mathbf q) \delta( \omega-\omega_{\mathbf q} ).$$

Use the $f$-sum rule to derive the single-mode relation. What changes if a continuum carries part of the weight?

Solution

The first moment of the assumed spectrum is

$$\int d\omega\, \omega S_n(\mathbf q,\omega) = \omega_{\mathbf q} S_n(\mathbf q).$$

The density $f$-sum rule gives

$$\omega_{\mathbf q} S_n(\mathbf q) = \frac{ \hbar q^2 }{ 2m }.$$

Hence

$$\hbar\omega_{\mathbf q} = \frac{ \hbar^2q^2 }{ 2mS_n(\mathbf q) }.$$

If a continuum carries weight, then

$$\frac{m_1}{m_0}$$

is the spectral mean frequency. It need not equal the position of a sharp mode, a threshold, or every excitation energy.

Related reading

• Fluctuation–Dissipation Theorem converts equilibrium scattering spectra to absorptive response with the correct thermal factor.
• Sum Rules derives the nested-commutator moments and their lattice counterparts.
• Spectral Functions distinguishes peaks, residues, continua, and linewidths across operator channels.
• Antiferromagnetism applies the spin structure factor to propagation vectors, domains, and magnetic Bragg peaks.

References

• P. B. Blakie, R. J. Ballagh, and C. W. Gardiner, “Theory of Coherent Bragg Spectroscopy of a Trapped Bose–Einstein Condensate”, Physical Review A 65, 033602 (2002).
• A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems, Dover (2003).
• R. P. Feynman, “Atomic Theory of the Two-Fluid Model of Liquid Helium”, Physical Review 94, 262–277 (1954).
• P. C. Hohenberg and W. F. Brinkman, “Sum Rules for the Frequency Spectrum of Linear Magnetic Chains”, Physical Review B 10, 128–131 (1974).
• S. W. Lovesey, Theory of Neutron Scattering from Condensed Matter, Clarendon Press (1984).
• P. Nozières and D. Pines, The Theory of Quantum Liquids, Westview Press (1999).
• L. Pitaevskii and S. Stringari, Bose–Einstein Condensation and Superfluidity, Oxford University Press (2016).
• G. L. Squires, Introduction to the Theory of Thermal Neutron Scattering, 3rd ed., Cambridge University Press (2012).
• J. Stenger, S. Inouye, A. P. Chikkatur, D. M. Stamper-Kurn, D. E. Pritchard, and W. Ketterle, “Bragg Spectroscopy of a Bose–Einstein Condensate”, Physical Review Letters 82, 4569–4573 (1999).
• L. Van Hove, “Correlations in Space and Time and Born Approximation Scattering in Systems of Interacting Particles”, Physical Review 95, 249–262 (1954).
• L. Van Hove, “Time-Dependent Correlations between Spins and Neutron Scattering in Ferromagnetic Crystals”, Physical Review 95, 1374–1384 (1954).