---
title: Cross-Phase Shift (XPS) in Optical Nonlinear Media
url: https://www.emergentmind.com/topics/cross-phase-shift-xps
type: topic
---

# Cross-Phase Shift (XPS) in Optical Nonlinear Media

Cross-phase shift (XPS) is the conditional phase shift impressed on one optical field by another through cross-phase modulation (XPM). In quantum language, a commonly used model is the cross-Kerr Hamiltonian
$$
\hat H=\hbar \chi \hat a_c^\dagger \hat a_c \hat a_p^\dagger \hat a_p,
$$
which gives a probe phase $\phi_p=\chi t n_c$ conditioned on the control photon number $n_c$; in semiclassical language, the same effect is written as a phase proportional to $\operatorname{Re}[\chi^{(3)}]$ and to the control intensity over the interaction region. Across the literature, XPS appears in far-off-resonant ladders, EIT and double-$\Lambda$ media, quantum memories, cavity-QED systems, thermal nonlocal media, epsilon-near-zero films, and post-selected weak-measurement settings [1505.03423, 2205.10007, 2205.13811, 2606.11516].

## 1. Formal definitions and microscopic descriptions

XPM denotes the third-order nonlinear modification of the refractive index seen by a probe due to a second field, while XPS denotes the resulting phase shift itself. In a homogeneous medium this is often written as
$$
\Delta \phi_p = k_p L \Delta n_p,
$$
with $\Delta n_p \propto \operatorname{Re}[\chi^{(3)}] I_s$ in the cross-Kerr limit. For a three-level ladder driven far off resonance, the Doppler-free unsaturated susceptibility can be written as
$$
\chi^{(3)} = - \frac{N \mu_1^2 \mu_2^2}{\epsilon_0 \hbar^3 \gamma_1^2 \gamma_2}\,
\frac{1}{(i+\delta_1)^2(i+\delta_2)},
$$
and the single-pass phase as
$$
\phi^{NC}=k_p n_2 \int_0^d I_c(z)\,dz,
$$
with $n_2=3\operatorname{Re}[\chi^{(3)}]/(2n_0^2\epsilon_0 c)$ [1505.03423].

In multilevel resonant media, XPS is more naturally expressed through density-matrix coherences. In the periodic tripod system, the probe susceptibility is expanded as
$$
\chi_p \approx \chi_p^{(1)}+\chi_{pp}^{(3)}|\Omega_p|^2+\chi_{pt}^{(3)}|\Omega_t|^2,
$$
so that in a uniform medium
$$
\Delta \phi_p \approx -\frac{\omega_1 L}{2c}\operatorname{Re}[\chi_{pt}^{(3)}]|\Omega_t|^2,
$$
whereas in a standing-wave medium the relevant quantities are the Fourier coefficients $\chi_{pt,0}^{(3)}$, $\chi_{pt,2}^{(3)}$, and $\chi_{pt,4}^{(3)}$ that enter the coupled forward and backward propagation channels [1012.2247].

A distinct but closely related representation arises in memory-based schemes. In gradient echo memory, the probe is stored as a spin coherence $\rho_{gs}$ and an off-resonant signal writes a phase through the AC Stark shift,
$$
\rho_{gs}(t)\rightarrow \rho_{gs}(t)e^{i\phi(t)}, \qquad 
\phi(T)=\int_0^T \delta_{AC}(t)\,dt,
$$
with $\delta_{AC}=|\Omega|^2/(4\Delta)$ in the far-detuned limit. In this picture the XPS is a phase on a collective ground-state coherence rather than on a propagating optical envelope [2205.10007].

## 2. Temporal structure, uniformity, and the distinction between peak and integrated phase

The temporal profile of XPS is platform-dependent. In EIT-assisted XPM driven by short pulses, a linear time-invariant description gives
$$
\phi(t)=\phi_0 n_{\mathrm{ph}}\cdot 2\tau \cdot e^{\tau_s^2/(2\tau^2)}\cdot e^{-t/\tau}
\left[1+\operatorname{erf}\!\left(\frac{t}{\sqrt{2}\tau_s}-\frac{\tau_s}{\sqrt{2}\tau}\right)\right],
$$
with $\Phi_{\mathrm{int}}=\int \phi(t)\,dt=\phi_0 n_{\mathrm{ph}}$. The EIT response time is
$$
\tau=\left[1+\frac{d_0}{4}\left(1-\frac{2\gamma}{\Gamma_{\mathrm{EIT}}}\right)\right]\frac{2}{\Gamma_{\mathrm{EIT}}},
$$
so narrowing $\Gamma_{\mathrm{EIT}}$ lengthens the decay even when the signal pulse is much shorter than the EIT response time. In this regime the phase rises on the signal timescale and decays on the EIT timescale; $\phi_{\mathrm{peak}}$ saturates for $\Gamma_{\mathrm{EIT}}\lesssim \Delta_s$, but $\Phi_{\mathrm{int}}$ continues to scale as $1/\Gamma_{\mathrm{EIT}}$ until limited by dephasing [1406.0245, 1506.07051].

The experimental realization of this short-pulse regime in laser-cooled $^{85}\mathrm{Rb}$ confirmed that with a $40\,\mathrm{ns}$ signal pulse and $\Gamma_{\mathrm{EIT}}=0.38\,\mathrm{MHz}$, the phase decay time was $\tau \approx 1.1\,\mu\mathrm{s}$. Narrowing the EIT window by an order of magnitude yielded nearly an order-of-magnitude increase in the integrated phase, while the peak phase increased by only about a factor of two before saturating. The integrated phase peaked near $\Gamma_{\mathrm{EIT}}=4\gamma$ [1506.07051].

Uniformity across a pulse is a separate issue from temporal duration. For two nonclassical pulses with different group velocities, an overtaking geometry can produce a nearly uniform XPM: once the faster pulse fully overtakes the slower pulse, the net phase becomes almost independent of the internal pulse shape. In the regularized instantaneous model the uniform phase after complete overtaking is
$$
\phi_i=\eta v_i \mathcal{V}_i/\Delta v,\qquad \Delta v=v_1-v_2,
$$
and strict unitarity requires $v_1V_1(z)=v_2V_2(z)$ so that $\phi_1=\phi_2$. If this condition is violated, additional Langevin or decoherence operators are required to preserve equal-time commutators, and the resulting mismatch severely degrades QND and controlled-phase-gate performance [1001.1893].

## 3. Resonantly enhanced atomic XPS in multilevel media

Atomic coherence engineering can make XPS strongly phase-dependent, spectrally selective, and channel-dependent. In the periodic tripod medium, a quasi-standing-wave control field produces a spatially periodic susceptibility, simultaneous transmitted and Bragg-reflected weak fields, and coupled-mode equations for forward and backward propagation. Under the representative parameter set $\Omega_c^+=4\,\mathrm{MHz}$, $\Omega_c^-=2\,\mathrm{MHz}$, $\Omega_{p0}=\Omega_{t0}=0.67\,\mathrm{MHz}$, $L=1.06\,\mathrm{mm}$, and $N\approx 1.3\times 10^{13}\,\mathrm{cm}^{-3}$, the trigger-induced XPS for the probe can reach order one radian over $L\approx 1\,\mathrm{mm}$. Detunings around $5.5\,\mathrm{MHz}$ and $6.4\,\mathrm{MHz}$ provide sizable transmission and reflection with appreciable $\Delta\phi_p^\pm$, while around $6.25\,\mathrm{MHz}$ the transmitted phase shift remains large and the reflected shift is negligible [1012.2247].

This tripod system also illustrates that population redistribution is not a small correction. The cross-Kerr term depends separately on $\sigma_{11}$ and $\sigma_{33}$, and population trapping near dressed-state resonances can push one of them close to unity within certain detuning ranges, dramatically enhancing or suppressing specific terms in $\chi_{pt}^{(3)}$. The paper’s analysis therefore treats population redistribution self-consistently over MHz-wide detuning ranges rather than fixing $\sigma_{11}=\sigma_{33}=1/2$ [1012.2247].

A different route to large low-light XPS uses a closed-loop double-$\Lambda$ system in cold $^{87}\mathrm{Rb}$, where the loop phase
$$
\Phi_{\mathrm{loop}}\equiv \phi_r=\phi_p-\phi_c+\phi_d-\phi_s
$$
acts as a control parameter. Under symmetric conditions, the probe output is
$$
\frac{\Omega_p(\alpha)}{\Omega_p(0)}
=\frac{1}{2}\left[1+e^{-i\phi_r}\right]
+\frac{1}{2}\left[1-e^{-i\phi_r}\right]e^{-i\alpha/(2\xi)},
$$
with $\xi=i+\Delta/\gamma_{31}$ for $|\Omega_c|=|\Omega_d|$. In a dark-SPOT ensemble with $\alpha_p\approx 46$–$52$, $|\Omega_c|=|\Omega_d|\approx 0.7\Gamma$, $\Delta=13\Gamma$, and $10\,\mu\mathrm{s}$ square pulses, the measured cross-phase shift induced by about 8 signal photons was $3.6\pm 1.0\,\mathrm{rad}$, corresponding to about $0.45\,\mathrm{rad}$ per photon under the reported conditions. The same experiment reported a $\pi$-level phase shift with two pulses both consisting of 8 photons, without cavities or tightly focusing of laser beams [1607.00818].

## 4. Cavity enhancement, stored excitations, and far-detuned optimization

Far-off-resonant cavities provide a quantitatively explicit XPS enhancement. Martínez-Rincón and Howell analyzed a ladder system inside a symmetric doubly resonant cavity and found
$$
\phi^C=\alpha(x,R)\phi^{NC},
\qquad
\alpha(x,R)=\frac{R e^{-x/2}(1+e^{-x})}{(1-Re^{-x})^2},
$$
with $x=OD/(1+\delta_1^2)$. At the optimal far-detuned operating point
$$
(\delta_1^*,\delta_2^*)\approx \left(\sqrt{OD\cdot F/\pi},-1\right),
$$
the maximum dimensionless phase factor satisfies
$$
g_{\max}\approx \frac{F}{4\pi},
$$
so that
$$
\phi_{\max}^C\approx \frac{F}{4\pi}\phi_{\max}.
$$
The same analysis identifies an off-resonance effective cooperativity
$$
\eta_{\mathrm{eff}}(\delta_1)=\frac{F}{\pi}\frac{OD}{1+\delta_1^2},
$$
with $\eta_{\mathrm{eff}}(\delta_1^*)=1$ at the optimum. At that point the control transmission is non-negligible but limited, with $OD_C(\delta_1^*)\approx \ln 4$ and $T_c\approx 25\%$; higher transmission can be obtained by moving farther off resonance at the cost of reduced XPS [1505.03423].

For the example ladder $5S_{1/2}\rightarrow 5P_{3/2}\rightarrow 5D_{5/2}$ in Rb, the same paper gives $\mu_2\approx 8.4\times 10^{-30}\,\mathrm{C\cdot m}$, $\gamma_1\approx 2\pi\times 6\,\mathrm{MHz}$, $\gamma_2\approx 2\pi\times 0.67\,\mathrm{MHz}$, $\lambda_c\approx 780.2\,\mathrm{nm}$, and $\lambda_p\approx 776\,\mathrm{nm}$. It reports $\phi_{\max}/I_0\approx 23\,\mathrm{rad}/(\mathrm{nW}/\mu\mathrm{m}^2)$ and $\bar\phi_{pp}\approx 88\,\mathrm{mrad}$ per atomic cross section in single pass, and $\phi_{\max}^C/I_0\approx (1.8\times F)\,\mathrm{rad}/(\mathrm{nW}/\mu\mathrm{m}^2)$ with $\bar\phi_{pp}^C\approx (7\times F)\,\mathrm{mrad}$ per atomic cross section in the cavity. With $I_0=1\,\mathrm{nW}$ focused to a $100\,\mu\mathrm{m}^2$ mode area and finesse $F\approx 175$, the predicted phase is about $\pi$ on the probe, corresponding to about $3.6\,\mathrm{mrad}$ per average intracavity control photon [1505.03423].

Memory-based and cavity-memory architectures aim to evade the pathologies of traveling-wave XPM by storing one field. In cold-atom GEM, the probe is stored as a long-lived ground-state coherence and a far-detuned signal imparts an AC Stark phase during storage; the reported inferred single-photon phase shift is $0.07\pm 0.02\,\mu\mathrm{rad}$, with excellent agreement with theoretical modelling and with degradation of memory efficiency at large phase shifts together with strategies to overcome that degradation [2205.10007]. In a high-finesse cavity containing an atomic quantum memory, a stored signal and a traversing control photon produced a conditional cross-phase shift of up to $\pi/3$ between the retrieved signal and control photons, a measured $\phi=0.4(1)\,\mathrm{rad}$ at $|\Delta|/2\pi=8\,\mathrm{MHz}$, time-postselected values up to $1.0(4)\,\mathrm{rad}$, and a positive concurrence $C=0.082(5)$ after correcting for detection and propagation losses [1512.02166].

A related cavity proposal using metastable xenon in a high-finesse cavity treats the XPS as a fourth-order dispersive shift in a three-level ladder. For $F\approx 6.0\times 10^4$, $L\approx 2.5\,\mathrm{mm}$, $w_0\approx 38\,\mu\mathrm{m}$, $\tau\approx 330\,\mathrm{ns}$, $\rho\approx 10^{10}\,\mathrm{cm}^{-3}$, and $\delta/2\pi\approx 10\,\mathrm{MHz}$, both perturbation theory and full matrix diagonalization predict a single-photon XPS of approximately $20\,\mathrm{mrad}$ [1403.6340].

## 5. Thermal and epsilon-near-zero material realizations

XPS is not confined to atomic resonances. In cylindrical lead glass, the dominant nonlinearity is thermal and highly nonlocal: absorption heats the medium, the temperature field modifies the refractive index through the thermo-optic coefficient, and the resulting broad parabolic waveguide guides a weak signal beam. In the pump-signal spatial-soliton geometry, synchronous propagation is obtained for
$$
w_s(0)=\frac{1}{\sqrt{\mu}}\,w_p(0),
$$
with $\mu=\lambda_p/\lambda_s$ and $\rho\approx 1$. The signal phase then contains a zero-order term proportional to pump power, and near the critical power it is approximately linear in the pump power. Experimentally, in a $60\,\mathrm{mm}$-long, $7.5\,\mathrm{mm}$-radius lead-glass rod with pump at $532\,\mathrm{nm}$ and signal at $790\,\mathrm{nm}$, the critical pump power was $P_c=260\,\mathrm{mW}$, the signal phase slope was about $61.5\,\mathrm{rad}/P_c$, and a change $\Delta P_p\approx 14\,\mathrm{mW}$ around $P_c$ produced approximately a $\pi$ phase shift of the signal soliton [1205.6106].

The thermal platform is intrinsically slow, since the response is steady-state and dominated by heat diffusion, but it produces very large controllable phase shifts with modest pump-power changes. The same work emphasizes that this regime is unsuitable for high-speed modulation compared with electronic Kerr media, even though it yields much larger phase-shift-per-mW in a centimeter-scale device [1205.6106].

At the opposite extreme, epsilon-near-zero ITO exhibits ultrafast nondegenerate XPS governed by hot-carrier redistribution. Near ENZ, small pump-induced changes in $\epsilon$ produce large $\Delta n$ through $\Delta n\approx \Delta\epsilon/(2n)$, and the paper reports that the probe placement at ENZ contributes strongly to the enhancement. In a $310\,\mathrm{nm}$ ITO film with $\lambda_{\mathrm{ENZ}}\approx 1250\,\mathrm{nm}$ and $n_{\mathrm{ENZ}}\approx 0.39+i0.39$, direct beam-deflection, spectral-shift, and Z-scan measurements were mutually consistent. With pump $\lambda_e=1250\,\mathrm{nm}$ and probe $\lambda_p=1200\,\mathrm{nm}$ at normal incidence and $I_e=100\,\mathrm{GW/cm^2}$, the measured phase shift was $\Delta\phi_{p,\mathrm{ITO}}=(0.44\pm 0.04)\pi\,\mathrm{rad}\approx 1.38\,\mathrm{rad}$ and $\Delta n_{\mathrm{eff}}=0.90\pm 0.08$, greater than the linear index near ENZ. The corresponding spectral shifts were $\Delta\lambda_p\approx +8.1\,\mathrm{nm}$ and $\Delta\lambda_p\approx -3.8\,\mathrm{nm}$, with sub-picosecond dynamics [2205.13811].

## 6. Single-photon conditioning, weak values, and limits for quantum logic

Recent work has separated deterministic XPS from conditioned peak XPS. A resonant narrowband photon transmitted through a cold atomic cloud and then post-selected in a short temporal detection mode produces a weak-value-amplified atomic excitation, and the probe phase is proportional to that conditioned excitation,
$$
\Delta\phi(t)=C\,P_e^w(t), \qquad C\approx 47\,\mu\mathrm{rad/excitation}.
$$
In the ideal thin-medium and $\delta$-post-selection limit the peak weak value is
$$
P_e^w(0)=1-e^{OD/2},
$$
which can be negative and can have magnitude greater than one. Experimentally, the maximum observed peak was $-148\pm 20\,\mu\mathrm{rad}$ at $OD=4.7\pm 0.4$, and at $OD=2.4\pm 0.1$ the conditioned peak XPS exceeded the best non-post-selected Gaussian single-photon-level pulse by a factor $6\pm 1$ [2606.11516].

This enhancement is explicitly conditional. The success probability is roughly the resonant transmission,
$$
P_{\mathrm{succ}}\approx e^{-OD},
$$
so increasing OD amplifies the conditioned peak while reducing success exponentially. The same work therefore concludes that the measured peak conditioned phase is orders of magnitude below $\pi$ and does not directly yield scalable cross-Kerr gates, even though it exposes a way around the usual bandwidth-duration trade-off for post-selected observables [2606.11516].

Several lines of work sharpen the distinction between large XPS and usable photonic logic. Traveling-wave EIT XPM is described as fundamentally limited by spontaneous-emission-induced phase noise and irreversible decoherence, whereas storing one mode in a spin wave or memory can decouple phase accumulation from optical loss in the probe channel [2205.10007, 1512.02166]. In the nonclassical-pulse overtaking model, an ideal controlled-phase gate requires equal cross-phase shifts for both pulses; if $\phi_1\neq \phi_2$, a Hamiltonian description fails, Langevin noise is required, and the concurrence for a target $\pi$ gate can be driven to zero when $|\phi_1-\phi_2|=\pi$ [1001.1893]. This suggests a sharp taxonomy: large peak XPS, large integrated XPS, large conditioned XPS, and large deterministic single-photon XPS are not interchangeable categories.

Across platforms, the central optimization variables recur. Large XPS favors steep dispersion or strong intracavity buildup, small mode area, large OD or finesse, and long effective interaction time; low-loss operation favors large detuning, uniform phase imprinting, and suppression of real excitation. The resulting trade-offs appear as $F/(4\pi)$ cavity enhancement with non-negligible transmission penalty in far-detuned ladders, $\Phi_{\mathrm{int}}\propto 1/\Gamma_{\mathrm{EIT}}$ with saturated peak phase in short-pulse EIT, microradian-level but low-noise single-photon shifts in GEM, radian-level conditional shifts with positive concurrence in cavity-memory systems, and large but intrinsically conditional weak-value peaks in post-selected resonant scattering [1505.03423, 1506.07051, 2205.10007, 1512.02166, 2606.11516].

Source: https://www.emergentmind.com/topics/cross-phase-shift-xps