Papers
Topics
Authors
Recent
Search
2000 character limit reached

Collision Quantum Optics

Updated 5 July 2026
  • Collision quantum optics is a field that studies quantum collisions such as Hong–Ou–Mandel events, repeated system-ancilla interactions, and structured mode scattering.
  • It uses both localized scattering and repeated-interaction models to explain phenomena like interference, exchange dynamics, and non-Markovian open-system behavior.
  • Experimental implementations span beam-splitter setups, quantum point contacts, and laser-pulse collisions, offering insights into optimal signal extraction and vacuum nonlinear effects.

Collision quantum optics can be understood as the body of quantum-optical work in which the central dynamical event is a collision: two excitations impinging on a beam splitter, a localized system interacting sequentially with time-bin modes, structured photonic states driving repeated open-system interactions, or several intense laser pulses overlapping in vacuum. Across these settings, the governing observables are collision-induced interference, exchange, repeated interaction, or vacuum-mediated signal-photon emission in a spacetime overlap region rather than isolated single-mode propagation. In the current literature, this umbrella includes Hong–Ou–Mandel-type beam-splitter physics, input-output and collision-model descriptions of open systems, electronic analogs on mesoscopic beam splitters, and strong-field QED processes in laser-pulse collisions (Ciccarello, 2017, Ferraro et al., 2016, Filippov et al., 2022, Gies et al., 2022).

1. Collision as a unifying quantum-optical motif

A first major strand treats a collision as a localized scattering event between a few excitations or modes. In this sense, the archetypal example is the Hong–Ou–Mandel (HOM) collision of two quanta at a 50:5050{:}50 beam splitter, together with electronic HOM analogs at a quantum point contact and laser-laser collision processes that generate signal photons from the nonlinear QED vacuum. A second strand treats a collision as a repeated interaction between a system and a sequence of ancillas or time bins, which is the collision-model interpretation of quantum input-output theory. A third strand generalizes the notion further to structured modes, such as vortex beams, or to repeated wave-packet scattering problems in which interparticle correlations are built and reshaped by multiple collisions. This plurality is explicit in the literature rather than imposed retroactively (Ciccarello, 2017, Cilluffo et al., 2020, Ivanov, 2011).

In the open-systems formulation, a collision model starts from an initial state

σ0=ρ0⊗(η⊗η⊗⋯ ),\sigma_0=\rho_0\otimes(\eta\otimes\eta\otimes\cdots),

and a one-step reduced map

ρn=E[ρn−1],\rho_n=\mathcal E[\rho_{n-1}],

with the continuous-time limit yielding a Lindblad equation under the standard assumptions of uncorrelated ancillas, no ancilla-ancilla interaction, and one collision per ancilla (Ciccarello, 2017). In waveguide and input-output language, the ancillas are coarse-grained temporal field modes, so the collision picture is not merely an analogy but a reformulation of familiar quantum-optical dynamics.

A complementary use of the term appears in strong-field QED. There, the “colliding objects” are not few-photon Fock states but intense laser pulses treated as classical backgrounds, while the signal is a single emitted photon from the nonlinear vacuum. The observable is then a directional and spectral signal-photon yield rather than a coincidence dip or a reduced density matrix (Gies et al., 2022, Sundqvist et al., 2023). Collision quantum optics is therefore not a single formalism; it is a family of collision-centered descriptions connected by interference, spacetime overlap, and mode-selective detection.

2. Beam-splitter collisions, HOM interference, and electronic analogs

In standard optical HOM theory, a two-photon Fock input

∣1⟩a∣1⟩b=a^†b^†∣0⟩|1\rangle_a|1\rangle_b=\hat a^\dagger \hat b^\dagger|0\rangle

transforms at a 50:5050{:}50 beam splitter so that the coincidence term c^†d^†\hat c^\dagger\hat d^\dagger cancels, leaving

a^†b^†∣0⟩→i2(c^†2+d^†2)∣0⟩.\hat a^\dagger \hat b^\dagger |0\rangle \rightarrow \frac{i}{2}\left(\hat c^{\dagger 2}+\hat d^{\dagger 2}\right)|0\rangle.

This is the standard bunching signature of two-photon indistinguishability and bosonic symmetry. A recent theoretical reinterpretation argues that the same HOM-type suppression can be reproduced in a selected coincidence subspace using attenuated coherent light, a 22.5∘22.5^\circ half-wave plate, a noninterfering Mach–Zehnder interferometer with synchronized oppositely scanned double-pass AOMs, polarization-basis selection by PBSs, and a final 50:5050{:}50 beam splitter. In that construction, the selected same-polarization pairs acquire opposite detunings ±δfj\pm\delta f_j, and the final joint amplitudes cancel in the coincidence channels while the local intensities remain uniform, ⟨Ik⟩=E02/2\langle I_k\rangle=E_0^2/2 (Ham, 2023).

The same work makes clear that this is not an unrestricted classical reproduction of HOM physics. The effect requires low mean photon number, independent sparse events, paired coherence from synchronized opposite AOM scanning, a fixed sum-phase relation, timing overlap, coincidence postselection, polarization filtering, and an effective 50%50\% loss of measurement events due to basis selection. It also assumes higher-order contamination is negligible at the ∼1%\sim1\% level. The paper presents this as a coherence-based reinterpretation of bunching in second-order detection, but it also states that the entanglement interpretation is nonstandard and may not be broadly accepted. A common misconception is therefore to treat all HOM-like coincidence suppression as equivalent to unconditional entanglement generation; the paper itself does not support that stronger claim (Ham, 2023).

Electronic quantum optics implements the same collision logic with fermionic excitations. In the integer quantum Hall setting, single-electron sources inject wave packets along ballistic edge channels toward a quantum point contact, and the collision is read out through zero-frequency current cross-correlations rather than photon coincidences. For identical electrons, the normalized HOM signal is

SHOM(δt)2SHBT=1−∣∫dτ  φ(τ) φ∗(τ+δt)∣2,\frac{\mathcal{S}_{HOM}(\delta t)}{2 \mathcal{S}_{HBT}} = 1- \left| \int d\tau \; \varphi(\tau) \, \varphi^*(\tau+\delta t)\right|^2,

so simultaneous arrival suppresses partition noise through Pauli antibunching. The same review shows that beyond the integer Hall regime, helical quantum spin Hall edges admit additional collision channels, including opposite-spin HOM dips and a three-electron collision with q(3)(0,0)=0q^{(3)}(0,0)=0, while Hall/superconductor hybrids replace ordinary electrons by Bogoliubov quasiparticles and render the HOM signal sensitive to superconducting phase differences (Ferraro et al., 2016).

A further caution comes from the classical limit of interacting-electron collisions. For two isolated electrons in a quantum Hall geometry scattered by a quadratic saddle-point beam splitter with unscreened Coulomb interaction, the guiding-center dynamics separates exactly into center-of-mass and relative coordinates, producing a universal phase diagram of deterministic bunching and antibunching. In that regime, classical Coulomb repulsion alone can mimic or obscure the suppression of same-output events that would otherwise be attributed to fermionic exchange. The relevant phase boundaries are

E− Φκ(E+/U)=±E+,E_-\,\Phi_\kappa(E_+/U)=\pm E_+,

and the interaction-induced saddle occurs at Ec=3U2E_c=\frac{3U}{2} (Pavlovska et al., 2022). In collision quantum optics with electrons, exchange and Coulomb dynamics must therefore be disentangled experimentally rather than assumed to coincide.

3. Collision models and the input-output structure of quantum optics

The collision-model formulation becomes exact and especially transparent in quantum optics when a propagating bosonic field is discretized into short time bins. Starting from the input operator

a^in(t)=12π∫dω a^(ω)e−iωt,\hat a_{\rm in}(t)=\frac{1}{\sqrt{2\pi}}\int d\omega\,\hat a(\omega)e^{-i\omega t},

one defines coarse-grained bin operators

α^n=1Δt∫tn−1tndt′ a^in(t′),\hat \alpha_n=\frac{1}{\sqrt{\Delta t}}\int_{t_{n-1}}^{t_n}dt'\,\hat a_{\rm in}(t'),

which satisfy bosonic commutation relations and play the role of ancillas. The corresponding collision Hamiltonian becomes

V^Sn=γΔt(b^ α^n†+b^† α^n),\hat V_{Sn}= \sqrt{\frac{\gamma}{\Delta t}}\left(\hat b\, \hat \alpha_n^\dag+\hat b^\dag\, \hat \alpha_n\right),

so the collision coupling scales as g=γ/Δtg=\sqrt{\gamma/\Delta t} and the rate g2Δt=γg^2\Delta t=\gamma remains finite in the continuous-time limit (Ciccarello, 2017). This explains why Lindblad dynamics emerges so naturally in memoryless collision models.

For vacuum input, the reduced dynamics is the standard damping master equation. For a coherent input, the discretized field is a product of weakly displaced ancillas, and the same construction yields the optical Bloch equation with a coherent drive term. For temporally correlated inputs, such as a single-photon wavepacket spread over many bins, the ancillas are initially correlated, so the next collision does not involve a fresh uncorrelated mode and the reduced dynamics can become non-Markovian (Ciccarello, 2017). In that sense, one of the most common collision-model assumptions in the abstract literature—noninteracting but initially correlated ancillas—has a direct physical origin in quantum-optical time-bin structure.

The same repeated-interaction picture extends to waveguide QED with many emitters and “giant” emitters, each coupled at several spatially separated points. In the negligible-delay regime

τN−τ1≪Δt≪γ−1,\tau_{\cal N}-\tau_1\ll \Delta t\ll \gamma^{-1},

one collision corresponds to one fresh time bin interacting collectively with all coupling points. The collision unitary contains both a collective emitter-bin term and an ordering-induced coherent Hamiltonian

H^vac=iγ2∑ν>ν′(A^ν′†A^ν−A^ν†A^ν′),\hat H_{\rm vac} = i\frac{\gamma}{2}\sum_{\nu>\nu'} \left( \hat{\cal A}_{\nu'}^\dag \hat{\cal A}_\nu - \hat{\cal A}_\nu^\dag \hat{\cal A}_{\nu'} \right),

which is of chiral origin and survives even when the geometric delays are negligible (Cilluffo et al., 2020). The resulting master equation unifies vacuum, coherent, thermal, squeezed, chiral, and bidirectional waveguide fields, and it exposes a notable possibility: if the collective couplings cancel, A^=0\hat{\cal A}=0 and A^′=0\hat{\cal A}'=0, the dissipative coupling vanishes while H^vac\hat H_{\rm vac} can remain nonzero, yielding decoherence-free Hamiltonian dynamics (Cilluffo et al., 2020).

Collision models also support measurement-conditioned dynamics. A non-Markovian trajectory algorithm based on a chain of field bins gives genuine photodetection and finite-local-oscillator homodyne unravelings for delayed coherent feedback and continuous-memory reservoirs. In that construction, one evolves the joint system-plus-memory state, measures the outgoing bin, discards it, and shifts the remaining bins forward. The key conceptual point is that in a non-Markovian setting the system-only conditioned state is generally mixed, because the unresolved field memory retains information about future recollisions (Whalen, 2019).

4. Correlated environments, exact embeddings, and programmable collision platforms

A major development in collision quantum optics is the treatment of correlated environments. For an environment state represented as a matrix-product state,

∣ψE⟩=∑i1,…,inB[1],i1B[2],i2⋯B[n],in∣i1i2…in⟩,\ket{\psi_E} = \sum_{i_1,\ldots,i_n} B^{[1],i_1}B^{[2],i_2}\cdots B^{[n],i_n}\ket{i_1 i_2 \ldots i_n},

the virtual bond indices can be reinterpreted as an auxiliary memory system. The enlarged state

R(kτ)=E[k] ⁣[R((k−1)τ)]R(k\tau)=\mathcal E^{[k]}\!\left[R((k-1)\tau)\right]

then evolves Markovianly on system plus bond space even when the physical environment induces non-Markovian reduced dynamics. This gives an exact tensor-network solution for repeated interactions with correlated reservoirs, and the associated time-convolution master equation makes the memory kernel explicitly dependent on environment correlations. At second order, the nonlocal term is controlled by the two-point operator-valued correlator

Cll′(∙)=trl,l′[∙(ϱl,l′−ϱl⊗ϱl′)].{\cal C}_{ll'} ( \bullet ) = {\rm tr}_{l,l'} [ \bullet (\varrho_{l,l'} - \varrho_{l} \otimes \varrho_{l'}) ].

The method is demonstrated for temporally ordered two-photon wavepackets, photonic cluster states, and AKLT spin chains (Filippov et al., 2022).

The exactness of collision models has been sharpened further by deriving them analytically from chain mapping. For a bath with cutoff ωc\omega_c, chain mapping is shown to be equivalent to a collision model with

Δt=πωc,\Delta t=\frac{\pi}{\omega_c},

both in the Markovian flat-spectrum case and in the structured non-Markovian case (Lacroix et al., 2024). This derivation identifies a previously overlooked error source: unfaithful sampling of the environment. If the time step satisfies Δt>π/ωc\Delta t>\pi/\omega_c, the environment is undersampled and aliasing in the bath correlation function produces a dominant error even when ordinary Trotter errors are small. For an Ohmic hard-cutoff bath, the sampling-error bound for ⟨σz⟩\langle \sigma_z\rangle scales as

ϵsamp≤2π2α[(ωcΔtπ)2−1]\epsilon_{\mathrm{samp}} \le 2\pi^2\alpha \left[ \left(\frac{\omega_c\Delta t}{\pi}\right)^2-1 \right]

when Δt≥π/ωc\Delta t\ge \pi/\omega_c (Lacroix et al., 2024). This is conceptually important: collision models are not exact merely because Δt\Delta t is “small”; they are exact only when the bath bandwidth is sampled faithfully.

Experimentally and numerically, collision-model ideas have been implemented in several photonic architectures. A bulk all-optical simulator based on two concatenated Sagnac interferometers realizes a stroboscopic collision model in which a system photon repeatedly collides with an environment mode and an ancilla photon witnesses information backflow through concurrence revivals. The memory parameter is the transmissivity TT of an attenuation stage: T=0T=0 is the most Markovian limit, T=1T=1 the maximal-memory case. The measured non-Markovianities are N=0.0044\mathcal N=0.0044 for T=0T=0, N=0.1442\mathcal N=0.1442 for T=0.209T=0.209, and N=0.3232\mathcal N=0.3232 for T=1T=1, with operation up to at least six collisions (Cuevas et al., 2018).

An integrated-photonics generalization uses cascaded Mach–Zehnder interferometers and SWAP operations to implement system-system, system-environment, and environment-environment collisions on chip. There, the tripartite mutual information and tripartite logarithmic negativity diagnose open-system scrambling, while the non-Markovianity is measured through trace-distance revivals

N(n)=∑k=1,D(k+1)>D(k)n−1[D(k+1)−D(k)].\mathcal N(n)= \sum_{\substack{k=1,\,D(k+1)>D(k)}}^{n-1} \left[D(k+1)-D(k)\right].

The work also introduces an observables-based compressed-sensing strategy for three-qubit tomography, using confidence intervals rather than exact equality constraints, and reports that Jeffreys intervals with α=13\alpha=\frac13 perform best for TMI prediction (Shang et al., 19 Jun 2025). This suggests that collision quantum optics is evolving from a conceptual framework into a programmable interferometric design language for open-system simulation.

5. Laser-pulse collisions and the nonlinear QED vacuum

In strong-field QED, collision quantum optics studies signal photons emitted when macroscopic laser pulses overlap in vacuum. The theoretical foundation is the one-loop Heisenberg–Euler interaction

Lint∼4F2+7G2,F=B2−E22,G=−E ⁣⋅ ⁣B,{\cal L}_{\rm int}\sim 4{\cal F}^2+7{\cal G}^2, \qquad {\cal F}=\frac{\mathbf B^2-\mathbf E^2}{2}, \qquad {\cal G}=-\mathbf E\!\cdot\!\mathbf B,

together with the vacuum-emission amplitude

S(p)(k)=⟨γ(p)(k)∣Γint[a^]∣0⟩.S_{(p)}(\mathbf k)= \langle \gamma_{(p)}(\mathbf k)|\Gamma_{\rm int}[\hat a]|0\rangle.

The experimentally relevant observable is the differential number of signal photons,

d3N(p)=d3k(2π)3∣S(p)(k)∣2,d^3N_{(p)}=\frac{d^3k}{(2\pi)^3}|S_{(p)}(\mathbf k)|^2,

resolved by frequency, direction, and sometimes polarization (Gies et al., 2022, Gies et al., 2021).

For two optical pulses in a pump–probe geometry, a central result is that maximal focusing does not maximize the discernible signal. The signal must be compared with the far-field laser background, and widening the probe focus often improves signal-to-background separation. For two 25 J25\,{\rm J}, 25 fs25\,{\rm fs}, 800 nm800\,{\rm nm} pulses colliding at ϑcol=160∘\vartheta_{\rm col}=160^\circ, the circular-probe optimum occurs around μ≈5.5\mu\approx5.5, giving

NSig≈5.839N_{\rm Sig}\approx 5.839

discernible photons per shot with background

NBg≈0.118.N_{\rm Bg}\approx 0.118.

For an elliptic probe, the configuration μx=5.5,μy=1\mu_x=5.5,\mu_y=1 yields

NSig≈11.633,NBg≈0.248.N_{\rm Sig}\approx 11.633, \qquad N_{\rm Bg}\approx 0.248.

In the polarization-flipped birefringence channel with purity P=10−10\mathcal P=10^{-10}, the same elliptic choice gives

NSig,⊥≈1.00,NBg,⊥≈0.03.N_{\rm Sig,\perp}\approx 1.00, \qquad N_{\rm Bg,\perp}\approx 0.03.

The general lesson is that collision-based vacuum signals are optimized by joint engineering of nonlinear conversion and far-field mode separation, not by peak field alone (Gies et al., 2022).

A complementary two-beam process is laser photon merging, where three ω\omega photons are merged into one 3ω3\omega photon,

ω+ω+ω→3ω.\omega+\omega+\omega\to 3\omega.

For two identical linearly polarized paraxial Gaussian beams, the emission is spectrally isolated by the factor

e−τ224(k0−3ω)2,e^{-\frac{\tau^2}{24}(k^0-3\omega)^2},

and the total yield exhibits an acute optimum at finite collision angle because exactly co-propagating beams give F=G=0{\cal F}={\cal G}=0. For two HPLS/ELI-NP 10-PW-class pulses with

ω=1.51 eV,τHM=24 fs,W=244 J,w0=λ,\omega=1.51\,{\rm eV},\quad \tau^{\rm HM}=24\,{\rm fs},\quad W=244\,{\rm J},\quad w_0=\lambda,

the numerically optimal angle is

ϑcollmax,num≈50.2∘,\vartheta_{\rm coll}^{\rm max,num}\approx 50.2^\circ,

with

Nmax3ω≈1.02N^{3\omega}_{\rm max}\approx 1.02

photons per shot, or about

$61.2$

photons per hour at $1/60$ Hz repetition. For a 1-PW-class CALA system, the estimate is about

$5.5$

merged photons per hour (Sundqvist et al., 2023).

Multi-color collisions enlarge the space of observable vacuum signatures by allowing explicit sum- and difference-frequency channels. A systematic channel-tracing method decomposes the signal into beam triples (i,j,l)(i,j,l) with overlap integrals Iijl\mathcal I_{ijl} and geometry factors g(p);ijlg_{(p);ijl}, and uses the approximate selection rule

k≃∣±ωi±ωj±ωl∣k \simeq |\pm\omega_i \pm\omega_j \pm\omega_l|

to identify promising channels. In a four-beam benchmark using ω0\omega_0, 2ω02\omega_0, and 4ω04\omega_0 pulses generated from a single 10 PW-class source, the total signal is about

Ntot≃5600N_{\rm tot}\simeq 5600

photons per shot, with well-separated peaks near ω0\omega_0, 2ω02\omega_0, 4ω04\omega_0, and 5ω05\omega_0. In selected off-axis angular regions, the yields are approximately

$62$

for a 2ω02\omega_0 channel,

$129.4$

for a 4ω04\omega_0 channel, and

$2.31$

and

$0.46$

for two distinct 5ω05\omega_0 channels, all with negligible laser background in the selected windows. Once the dominant microscopic channel is identified, spectator beams can be removed and the participating beam energies reoptimized; for the 4ω04\omega_0 inelastic channel this raises the predicted yield to about

$186$

photons per shot (Gies et al., 2021).

A weaker-field analytical treatment of head-on light-beam collisions reaches a related but distinct conclusion: the nonlinear vacuum source oscillates at 3ω3\omega, and stationary-phase analysis predicts secondary waves associated with conical directions obeying

α=arccos⁡(1/3).\alpha=\arccos(1/3).

The same framework shows that the time-averaged radiation-pressure correction in interferometers such as LIGO is fantastically small, of order 10−3310^{-33} for representative parameters, so the effect is conceptually real but presently negligible for gravitational-wave detection (Hacyan, 2020).

6. Structured-mode collisions, repeated wave-packet scattering, and broadened scope

Collision quantum optics also includes work in which the colliding objects are structured beams or repeated motional wave packets rather than ordinary optical modes. Ivanov showed that elastic scattering of a Bessel vortex beam with a counterpropagating plane wave produces a final state that is naturally entangled in both orbital helicities (m1,m2)(m_1,m_2) and cone parameters (ϰ1,ϰ2)(\varkappa_1,\varkappa_2). The triple-twisted amplitude depends jointly on the two outgoing OAM quantum numbers and the two opening-angle variables, and the allowed (ϰ1,ϰ2)(\varkappa_1,\varkappa_2) region is restricted by

∣ϰ1−ϰ2∣≤ϰ~≤ϰ1+ϰ2.|\varkappa_1-\varkappa_2|\le \tilde\varkappa \le \varkappa_1+\varkappa_2.

This makes vortex entanglement a kinematical consequence of beam-beam scattering rather than a nonlinear-medium effect, and the mechanism is presented as species-independent, applying to eγe\gamma, e+e−e^+e^-, epep, and related collisions (Ivanov, 2011).

Repeated wave-packet collisions provide a second generalized setting. For two hard-core particles of masses mx≪mym_x\ll m_y in one dimension with a reflecting wall, a decomposition into non-entangling channels yields an analytically controlled solution of the many-collision problem. The key width-matching condition

mxσ0x2=myσ0y2m_x\sigma_{0x}^2=m_y\sigma_{0y}^2

ensures that within each channel the collisions preserve product-Gaussian form. The full state is then reconstructed from a classical distribution over channel centers, and the analysis reveals an unexpected entanglement history: repeated collisions first increase entanglement, but later the channel distribution refactorizes and the entanglement disappears completely when the light particle becomes too slow to catch the heavy one. The maximal number of collisions is

nmax⁡≈π4ϵ,ϵ=mxmy,n_{\max}\approx \frac{\pi}{4\epsilon}, \qquad \epsilon=\sqrt{\frac{m_x}{m_y}},

and the mixed Gaussian coefficient AnxyA_n^{xy} vanishes both initially and at the end of the collision cascade (Hahn et al., 2011). This suggests that collision-generated correlations need not be monotonic even in fully coherent unitary dynamics.

The conceptual reach of the term has also begun to expand beyond standard open-system and scattering theory. A recent graphics-oriented formulation treats layered coherent scattering as a sequence of symmetry-constrained unitary collisions between quantized light modes and localized surface excitations, using a three-mode collision unitary in the single-excitation sector,

U1=(ratacatar),U_1= \begin{pmatrix} r & a & t\\ a & c & a\\ t & a & r \end{pmatrix},

together with repeated collisions across a stack of layers (Ferreira et al., 29 Jun 2026). The same work explicitly states that this is best read as a custom graphics-oriented adaptation rather than a standard master-equation or input-output derivation. It marks a boundary case: collision language is being exported into coherent-scattering problems that are quantum-optics-adjacent but not part of the conventional collision-model canon.

Taken together, these extensions show that collision quantum optics is not exhausted by HOM interferometry or open-system time-bin models. It also provides a framework for understanding how mode structure, repeated scattering, and spatiotemporal overlap produce entanglement, memory, and observable signal channels across photonic, electronic, and vacuum-nonlinear regimes.

Topic to Video (Beta)

No one has generated a video about this topic yet.

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to Collision Quantum Optics.