Superexchange Interactions in Magnetic Systems

• Superexchange interactions are indirect magnetic couplings mediated by virtual charge fluctuations across bridging atoms, defining effective spin Hamiltonians in insulating and engineered systems.
• The mechanism involves multiple exchange channels, including ferromagnetic, bond-directional, and anisotropic terms arising from orbital geometry, spin-orbit coupling, and lattice dynamics.
• Contemporary studies leverage advanced quantum models and external fields to engineer superexchange, enabling tunable magnetic interactions in quantum dots, optical lattices, and polaritonic materials.

Superexchange interactions are indirect magnetic couplings produced by virtual charge motion through high-energy intermediate states. In the canonical insulating case, localized moments on magnetic sites couple through nonmagnetic ligands, so that charge fluctuations are frozen in the ground state but still generate an effective spin Hamiltonian after downfolding. Contemporary work shows that this mechanism is broader than the textbook ligand-mediated antiferromagnetic exchange of simple Mott insulators: superexchange can be ferromagnetic, bond-directional, multi-anion, controlled by nominally nonmagnetic cations, reshaped by spin-orbit coupling, and engineered in quantum dots, driven optical lattices, and trapped-ion polariton systems (Furuya et al., 2021, Huang et al., 2019, Chang et al., 2019, Ivanov et al., 2014).

1. Microscopic formulation and effective-spin reduction

At the formal level, superexchange is a low-energy consequence of eliminating virtual charge excitations. In insulating $d$-$p$-$d$ models this frequently appears at fourth order in hopping, whereas in already localized manifolds—such as polaritonic Mott states or detuned charge states of quantum dots—it can appear as a second-order process in the residual hopping. A compact example is the single-ligand antiferromagnetic model of a Mott insulator in a DC electric field, where the effective Hamiltonian takes the form $$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$, with $J_{\rm A}(\mathbf 0)>0$ under the conditions $\Delta_j(\mathbf 0)\ge 0$ and $U_d>U_p>0$ (Furuya et al., 2021).

In multi-orbital charge-transfer systems, the exchange constant depends explicitly on the virtual hopping geometry. For one-dimensional corner-shared cuprates, the effective antiferromagnetic scale is written as

$$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$

so that both Cu–O hoppings and the charge-transfer gap enter directly (Li et al., 2022). This structure already makes clear that superexchange is not a purely geometric label; it is a perturbative amplitude built from orbital-resolved transfer integrals and interaction energies.

The resulting spin model is not unique across the literature, and even the sign convention is model dependent. In graphene trimers the effective Heisenberg model is written as ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$, so $J<0$ denotes antiferromagnetic coupling (Crook et al., 2018). In EuZn$p$0P$p$1, by contrast, the effective model is $p$2, so positive $p$3 favors parallel alignment (Singh et al., 2023). For this reason, the physical content of a reported “positive” or “negative” exchange constant must always be read together with the Hamiltonian convention.

2. Geometry, bridge chemistry, and the sign of exchange

The sign and magnitude of superexchange are controlled by the geometry and electronic activity of the bridge. In EuZn$p$4P$p$5, the intralayer Eu–P–Eu path has bond angle $p$6 and is assigned a ferromagnetic sign, whereas the interlayer Eu–P–P–Eu path has Eu–P–P angle $p$7 and is assigned an antiferromagnetic sign. DFT mapping gives $p$8 meV and $p$9 meV, and Monte Carlo with these two couplings yields $d$0 K, close to the experimental $d$1 K (Singh et al., 2023). This is an “extended” superexchange in the paper’s terminology, because the interlayer path involves two phosphorus atoms rather than a single anion.

Bridge identity can be as important as bond angle. In the double perovskites Sr$d$2CuTeO$d$3 and Sr$d$4CuWO$d$5, formally nonmagnetic Te$d$6 and W$d$7 produce qualitatively different super-superexchange networks. Quantum-chemistry calculations give $d$8 meV and $d$9 meV for Sr$$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$0CuTeO$$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$1, but $$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$2 meV and $$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$3 meV for Sr$$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$4CuWO$$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$5. The reason is that low-lying empty W $$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$6 states strongly enhance the $$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$7 Cu–O–W–O–Cu path, whereas in the Te compound the dominant $$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$8 is mainly a Cu–O–O–Cu process and the Te-centered diagonal path remains weak (Katukuri et al., 2019). This is a direct demonstration that nominally nonmagnetic cations can be active elements of the exchange bridge rather than passive spacers.

Near-$$\mathcal H_{\rm A}^{\rm eff}=J_{\rm A}(\mathbf E)\,\mathbf S_0\cdot\mathbf S_1-h(S_0^z+S_1^z)$$9 geometries do not force a unique outcome. In CaMnCrSbO$J_{\rm A}(\mathbf 0)>0$0, the nearest-neighbor Mn–O–Mn and Mn–O–Cr channels are all antiferromagnetic but weak, with fitted values such as $J_{\rm A}(\mathbf 0)>0$1 meV and $J_{\rm A}(\mathbf 0)>0$2 meV for Mn–Cr exchange, while the next-nearest-neighbor Cr–O–O–Cr super-superexchange is ferromagnetic with $J_{\rm A}(\mathbf 0)>0$3 meV. The paper argues that the correct sign of this Cr–O–O–Cr term requires a four-site model appropriate to doubly occupied oxygen $J_{\rm A}(\mathbf 0)>0$4 orbitals rather than the usual hole-model formula, which would give the wrong sign (Dhawan et al., 2022).

The strongest departure from textbook Anderson superexchange is the explicit ferromagnetic kinetic channel identified for diamagnetic metal bridges. In a three-orbital model with two localized magnetic orbitals and one bridge orbital, the exchange contains a third-order term

$J_{\rm A}(\mathbf 0)>0$5

which is ferromagnetic when $J_{\rm A}(\mathbf 0)>0$6. In the Fe$J_{\rm A}(\mathbf 0)>0$7–Co$J_{\rm A}(\mathbf 0)>0$8–Fe$J_{\rm A}(\mathbf 0)>0$9 complex analyzed in detail, the decomposition yields $\Delta_j(\mathbf 0)\ge 0$0 meV, $\Delta_j(\mathbf 0)\ge 0$1 meV, $\Delta_j(\mathbf 0)\ge 0$2 meV, $\Delta_j(\mathbf 0)\ge 0$3 meV, $\Delta_j(\mathbf 0)\ge 0$4 meV, and total $\Delta_j(\mathbf 0)\ge 0$5 meV, so the large ferromagnetic coupling is driven predominantly by the cyclic kinetic term rather than by conventional potential exchange (Huang et al., 2019). A common misconception is therefore that kinetic exchange through diamagnetic bridges is generically antiferromagnetic; this paper shows that the conclusion depends on how explicitly the bridge orbitals and magnetic-center transfer $\Delta_j(\mathbf 0)\ge 0$6 are retained.

3. Spin-orbit-entangled and anisotropic superexchange

When spin-orbit coupling is large, superexchange is not exhausted by isotropic Heisenberg exchange. In the $\Delta_j(\mathbf 0)\ge 0$7 $\Delta_j(\mathbf 0)\ge 0$8 problem treated in the $\Delta_j(\mathbf 0)\ge 0$9-$U_d>U_p>0$0 coupling scheme, the local basis is reorganized into $U_d>U_p>0$1 and $U_d>U_p>0$2 spin-orbit-entangled orbitals. Two virtual channels then appear naturally: $U_d>U_p>0$3-to-$U_d>U_p>0$4 processes generate a ferromagnetic Ising interaction identified with a quantum-compass term, whereas $U_d>U_p>0$5-to-$U_d>U_p>0$6 processes generate antiferromagnetic Heisenberg exchange. The paper further argues that increasing the spin-orbit coupling $U_d>U_p>0$7 suppresses the ferromagnetic Ising contribution, so the dominant exchange changes from ferromagnetic Ising type to antiferromagnetic Heisenberg type as $U_d>U_p>0$8 increases (Matsuura et al., 2014).

A second development is the explicit role of spin-orbit coupling on the nonmagnetic bridge itself. In a multi-orbital Hubbard model with SOC on both magnetic cations and nonmagnetic anions, local SOC rotates the hopping into the form $U_d>U_p>0$9, so anisotropic exchange emerges already at the level of the effective hopping matrices. The resulting fourth-order superexchange contains the isotropic Heisenberg term, a Dzyaloshinskii–Moriya vector $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$0, a symmetric anisotropic tensor $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$1, and single-ion anisotropy; in the weak-SOC limit, $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$2 is linear in SOC and $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$3 quadratic. The central claim is that SOC on nonmagnetic anions can induce these anisotropies on an equal footing with SOC on magnetic ions (Chang et al., 2019).

This shift in viewpoint is significant for heavy-ligand compounds. It implies that ligand SOC is not merely a correction to an otherwise cation-controlled exchange problem, but can be a primary source of bond-directional and antisymmetric exchange. A plausible implication is that the conventional separation between “exchange geometry” and “single-ion spin-orbit physics” becomes inadequate once the bridge orbitals themselves are spin-orbit active.

4. Exchange topology, dimensionality, and materials realization

Superexchange determines not only the sign of a pairwise coupling but also the dimensionality of the magnetic network. In tetragonal multiferroic $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$4, reducing the particle size weakens the ferroelectric distortion and shortens the long apical Fe–O bond from $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$5 Å at $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$6 nm to $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$7 Å at $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$8 nm, while tetragonality decreases from $$J=\frac{t_{pd,1}^2 t_{pd,2}^2}{\Delta_{dp}^2}\left(\frac{1}{\Delta_{dp}+U_p/2}+\frac{1}{U_d}\right),$$9 to ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$0. The authors interpret this as restoring the Fe–O–Fe exchange path along the ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$1 axis, producing a crossover from essentially ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$2 antiferromagnetic interactions in bulk-like particles to ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$3 antiferromagnetic interactions in nanoparticles. Experimentally, ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$4 rises from ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$5 K to ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$6 K, opposite to ordinary finite-size suppression in non-multiferroic antiferromagnets (Upadhyay et al., 2014).

In iron pnictides, the dominant exchange path can be shifted away from nearest neighbors. For LaFePO, first-principles calculations give ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$7 meV/${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$8 and ${\cal H}=-(J/2)\sum \vec S_i\cdot\vec S_j$9 meV/$J<0$0, with the next-nearest-neighbor $J<0$1 attributed to P $J<0$2-bridged superexchange on the Fe square lattice. The small $J<0$3 and dominant $J<0$4 stabilize the collinear stripe state and are tied to the same pnicogen-mediated exchange picture previously advanced for LaFeAsO (Lu et al., 2010). In low-symmetry two-dimensional magnets, the same logic generalizes from “which sign?” to “which Hamiltonian?”: in monolayer Cr$J<0$5Te$J<0$6, DFT plus Wannier path counting yields six distinct nearest-neighbor exchange constants rather than a single $J<0$7, with $J<0$8 meV the strongest term and a calculated $J<0$9 K close to the reported $p$00 K (Li et al., 2023).

Carbon-based systems illustrate how interface symmetry reshapes superexchange. In zigzag graphene nanoribbons embedded in graphane, symmetric C–CH/C–CH interfaces generate strong, long-range interedge antiferromagnetic superexchange for all $p$01, with $p$02 rising from $p$03 meV/unit cell at $p$04 to $p$05 meV at $p$06. In zigzag ribbons embedded in h-BN, by contrast, asymmetric C–B and C–N edges make the interedge coupling short-ranged and weak, so the ground state is nonmagnetic except at $p$07 and $p$08, where only a weakly stabilized half-semimetallic state appears with $p$09 and $p$10 meV/unit cell (Kim et al., 2015). In graphene trimers, the interpretation is even less canonical: Cr and Mn trimers are antiferromagnetic-like at $p$11 but ferromagnetic at $p$12 and $p$13, and the authors describe the mechanism as “RKKY-like super-exchange” because it combines substitutional local bonding with host-mediated, metallicity-dependent, distance-sensitive coupling rather than fitting a purely local superexchange picture (Crook et al., 2018).

Fluoride polymorphs show a related control of exchange topology by structural reorganization. In AgMF$p$14 ($p$15 Ni, Cu), quasi-linear Ag–F–M bridges become the dominant antiferromagnetic pathways once the lattice connectivity departs from stacked AgF$p$16-like layers. Low-pressure AgCuF$p$17 remains mainly layer-dominated, with strong Ag–Ag and Cu–Cu superexchange, but compression yields $p$18 meV in AgCuF$p$19-HP. AgNiF$p$20 likewise develops dominant mixed-chain couplings, with $p$21 meV in the low-pressure phase and $p$22 meV in the high-pressure phase (Domański et al., 2021).

5. External fields, lattice dynamics, and nonequilibrium control

Because superexchange depends on virtual excitation energies, it is electrically tunable even in insulators. A general DC-field theory shows that static electric fields modify the site-energy differences $p$23 entering the virtual processes, thereby changing both antiferromagnetic and ferromagnetic superexchange. On the square lattice, this permits control of $p$24 and can even split the two diagonal couplings into $p$25, effectively deforming a square-lattice $p$26-$p$27 magnet into a triangular one. On a spin chain, a field-induced bond alternation $p$28 produces singlet-dimer or Haldane-dimer order; for the Heisenberg antiferromagnetic chain the induced gap scales as $p$29. The paper estimates observable field scales of order $p$30 for inorganic Mott insulators and $p$31 for organic ones, and gives $p$32 K for $p$33 at $p$34, about $p$35 of $p$36 K (Furuya et al., 2021).

Lattice fluctuations can renormalize the same virtual processes in the opposite direction. In a four-orbital Hubbard–Su–Schrieffer–Heeger model for one-dimensional cuprates, bond-stretching oxygen phonons modulate the two Cu–O hoppings oppositely, so the exchange is renormalized as $p$37. Exact diagonalization and determinant quantum Monte Carlo show that increasing the electron-phonon coupling suppresses the effective superexchange, softens the spin spectrum, and enhances the uniform susceptibility. A narrow dimerized regime exists only just below a critical coupling $p$38; beyond that, the linear SSH model becomes unstable because the effective hopping changes sign (Li et al., 2022). This directly contradicts the common expectation that phonons necessarily enhance exchange by reducing an effective $p$39; in the charge-transfer geometry studied here, the dominant effect is instead the reduction of the hopping product entering superexchange.

Time-periodic driving provides a nonequilibrium control knob. In AC-driven optical double wells, a magnetic-field gradient $p$40 suppresses static superexchange, but lattice modulation restores it resonantly at $p$41. In the symmetric case $p$42, the photon-assisted transverse exchange is

$p$43

while the longitudinal term is

$p$44

This produces an effective XXZ Hamiltonian with independently tunable transverse and Ising couplings, and the experiment reports a one-photon superexchange resonance at $p$45 with $p$46 (Chen et al., 2011).

6. Mesoscopic and synthetic realizations

In semiconductor triple quantum dots, superexchange becomes a directly gate-defined long-range resource. For a two-electron linear triple dot with an empty mediator, full configuration interaction finds that the outer-spin superexchange $p$47 can be non-monotonic and can switch sign as a function of middle-dot detuning when the outer-dot detunings are leveled. The same study shows that increasing left-right detuning strengthens an originally positive $p$48 and weakens an originally negative one, and that simplified Hubbard descriptions miss this behavior unless long-range Coulomb exchange $p$49 is included (Chan et al., 2022).

Mediator occupancy adds a second control axis. Configuration-interaction calculations for a larger central dot show that an empty mediator produces at most about a $p$50 enhancement of the remote coupling, whereas loading two electrons into the mediator can enhance the outer-dot exchange by more than two orders of magnitude at small angle. In the linear geometry, the coupling remains above $p$51 out to roughly $p$52 nm, compared with about $p$53 nm for the direct double-dot reference (Deng et al., 2020). With a four-electron mediator, the sign itself becomes occupancy and field dependent: a two-electron singlet mediator gives positive exchange, a four-electron singlet mediator negative exchange, and a four-electron triplet mediator under a non-uniform magnetic field can be tuned through zero as $p$54 is increased relative to $p$55; leakage from the logical subspace is estimated to fall below $p$56 when $p$57 T (Chan et al., 2022).

Quantum-dot superexchange can also display interference. In a triple dot with two distinct virtual intermediate states, $p$58 and $p$59, the effective singlet amplitude is

$p$60

whereas the triplet amplitude lacks the $p$61 path. Destructive interference at $p$62 gives $p$63, producing a dark state and a “superexchange blockade” of current (Sánchez et al., 2013). This is a mesoscopic analogue of path-interference control over an otherwise conventional virtual-exchange mechanism.

Synthetic many-body platforms realize the same logic with different microscopic carriers. In a linear trapped-ion crystal, a V-type Jaynes–Cummings–Hubbard model creates Mott-localized ion-phonon polaritons, and second-order phonon hopping generates effective spin models. At one excitation per site the low-energy theory is an XXZ spin-$p$64 model; at two excitations per site it becomes a spin-1 Heisenberg-type model with additional anisotropies. The exchange strengths are tunable through trap frequencies, laser intensities, and detuning (Ivanov et al., 2014).

Taken together, these results suggest that “superexchange interaction” is best understood not as a single fixed antiferromagnetic mechanism, but as a family of virtual-exchange processes whose sign, anisotropy, range, and dimensionality are set by the detailed structure of the intermediate manifold—ligand identity, bridge multiplicity, orbital symmetry, spin-orbit entanglement, lattice dynamics, and external control parameters.