Automated experimental design for high-probability entanglement generation
Carlos Ruiz-Gonzalez, Mario Krenn, Xuemei Gu · 2026
Read Automated experimental design for high-probability entanglement generation by Carlos Ruiz-Gonzalez, Mario Krenn, Xuemei Gu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-02721.txt.
Introduction
Carlos Ruiz-Gonzalez, Mario Krenn, and Xuemei Gu build an algorithm that designs photon experiments to make entanglement with high success odds, not just high quality. Entangled photons are a workhorse of quantum technology. Sources that make photon pairs are probabilistic, and chasing perfect fidelity can make success vanishingly rare, a trade-off often ignored.
Standard treatments keep only the single-pair term and drop multi-pair emissions. The new design algorithm instead accounts for those higher-order events, explores different layout topologies under hardware limits, and tunes source settings to shrink the unwanted terms, or even use them. For widely used target states, including heralded Bell, W, and NOON states, the found designs beat prior proposals. The result points to more efficient photonic setups.
Catalog record
| CREATOR | Carlos Ruiz-Gonzalez, Mario Krenn, Xuemei Gu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.02721
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Automated experimental design for high-probability entanglement generation Carlos Ruiz-Gonzalez, Mario Krenn, Xuemei Gu Abstract Entangled photons are widely used in quantum technologies. Many photonic experiments generate them with probabilistic photon-pair sources that can be modeled as squeeze operators. In practice, these sources are usually treated in the low-gain (perturbative) regime, keeping only the leading single-pair term and neglecting higher-order multi-pair emission events. In pursuit of fidelity, the probability of successful entanglement generation can become extremely small, a tradeoff often ignored. Here we develop an automated design algorithm for quantum experiments to optimize both fidelity and success probability while accounting for higher-order multi-pair emissions. Our discovery algorithm explores different design topologies subject to varying hardware constraints. It optimizes the source parameters to reduce undesired higher-order terms or even benefit from them. The experiments presented outperform previous proposals for widely used states, including heralded Bell states, W states, and NOON states, paving the way for more efficient photonic technologies. Introduction — Entanglement is a crucial resource in quantum technologies 25 , enabling quantum computing 7 ; 1 ; 43 , sensing 27 ; 40 , and communications 3 ; 57 . Photonics is a promising platform for these advances 39 ; 52 : photons can travel long distances with low decoherence, often without cryogenic cooling 49 ; 36 , and benefit from mature technologies such as optical fiber and integrated circuits 37 ; 42 ; 41 . To generate entangled photons, many setups combine probabilistic sources, such as spontaneous parametric down-conversion (SPDC), with linear elements and different detection schemes 45 ; 17 . When pumped with strong fields, SPDC sources can produce a high number of photon pairs 13 . Yet, most experiments operate in the perturbative low-gain regime. The pump is so weak that multi-pair emission events are negligible, and each source is approximated by a single pair-creation term. Successful state generation is then identified by specific detector-click patterns 17 . However, as the reported success probability is conditional on those detection outcomes, a fundamental tradeoff is overlooked. As the pump decreases to preserve fidelity, the probability of obtaining the required detection pattern becomes extremely small. Here, we model SPDC sources as full squeeze operators 44 ; 9 ; 12 ; 20 , revealing that the photon statistics neglected by the perturbative approximation are central for efficient entanglement generation. This turns experimental design into a joint optimization of two competing properties: fidelity and success probability. To improve the tradeoff, we engineer interference between higher-order contributions, reducing undesired emissions. Interestingly, we see that higher-order events can enhance the experiments’ performance when using ancillary photons. We first revisit the generation of GHZ states to illustrate how the tradeoff emerges beyond the low-gain approximation. We then present our algorithm to optimize the experiments’ parameters and configurations, maximizing the success probability for a range of target fidelities. Finally, we apply the methods to produce a 4-qubit W state, a heralded Bell state, and N=3 and N=4 N00N states, surpassing previous proposals 18 ; 46 ; 45 . These are widely used states with different entanglement structures, detection schemes, and photon statistics. By moving beyond the low-gain approximation, our work sheds new light in probabilistic entanglement generation. Figure 1: Generating a 4-qubit GHZ state. (a) We replicate an entanglement generation experiment with SPDC crystals and threshold detectors 33 . When all detectors click, the projected state is a superposition of horizontally and vertically polarized photons from the blue (I & II) and orange (III & IV) sources, respectively. With two Fock modes, we encode the photons as \left|\text{H}\right\rangle\!=\!\left|1,0\right\rangle and \left|\text{V}\right\rangle\!=\!\left|0,1\right\rangle . (b) The SPDC crystals act as squeeze operators on 4 different optical paths, initialized at vacuum with 2 polarization modes: \left|0,0\right\rangle . Splitting each path into 2 modes, we see the sources commute, as they act on different modes. (c) Increasing the common pump r , we transcend the low-gain approximation and reveal the tradeoff between fidelity and success probability. Here and in following figures, the shown counts per second assume the crystals are pumped by a laser with 10ns repetition rate. Beyond the low-gain regime — The setup of Fig. 1 , introduced in the low-gain regime 33 , generates a 4-qubit GHZ state, (\left|\text{HHHH}\right\rangle+\left|\text{VVVV}\right\rangle)/\sqrt{2} , from four SPDC sources. Since the photodetectors ignore polarization, we cannot know which sources fired when each detector clicks. This indistinguishability of emission events is the basis of path identity experiments 58 ; 54 ; 38 ; 23 ; 4 ; 53 ; 26 ; 6 . In Fig. 1 a, with a common pump r\ll 1 and phase \theta=0 , each source can be approximated as S_{2}(r,\theta)=\exp(r(e^{-i\theta}ab-e^{i\theta}a^{\dagger}b^{\dagger}))\approx(1+rab-ra^{\dagger}b^{\dagger}) , where a ( a^{\dagger} ) and b ( b^{\dagger} ) are annihilation (creation) operators acting in two different modes. The resulting state is, approximately, \displaystyle\left|\psi\right\rangle= \displaystyle S_{\text{I}}S_{\text{II}}S_{\text{III}}S_{\text{IV}}\left|vac\right\rangle \displaystyle\approx \displaystyle(1+rh_{1}h_{3}-rh_{1}^{\dagger}h_{3}^{\dagger})(1+rh_{2}h_{4}-rh_{2}^{\dagger}h_{4}^{\dagger}) \displaystyle(1+rv_{1}v_{2}-rv_{1}^{\dagger}v_{2}^{\dagger})(1+rv_{3}v_{4}-rv_{3}^{\dagger}v_{4}^{\dagger})\left|vac\right\rangle, (1) where h_{j}^{\dagger} ( v_{j}^{\dagger} ) is the creation operator of horizontally (vertically) polarized photons in the j- th optical path. When the four threshold detectors click, the final state is \displaystyle|\tilde{\psi}\rangle=\> \displaystyle r^{2}(h_{1}^{\dagger}h_{2}^{\dagger}h_{3}^{\dagger}h_{4}^{\dagger}+v_{1}^{\dagger}v_{2}^{\dagger}v_{3}^{\dagger}v_{4}^{\dagger})\left|vac\right\rangle \displaystyle-r^{3}(h_{1}^{\dagger}h_{2}^{\dagger}h_{3}^{\dagger}h_{4}^{\dagger}v_{1}^{\dagger}v_{2}^{\dagger}+\dots)\left|vac\right\rangle+\dots \displaystyle\approx\> \displaystyle r^{2}(\left|\text{HHHH}\right\rangle+\left|\text{VVVV}\right\rangle). (2) Hence, in the low-gain approximation, terms with extra photons are neglected, annihilation operators can be omitted, and the fidelity is assumed optimal. What the approximation obscures is how unlikely successful detection is, since the dominant terms with unpopulated paths are left out. Beyond this low-gain regime, when the pump grows, so does the success probability. At the same time, as the weight of high-order terms increases, the fidelity drops. This tradeoff is shown in Fig. 1 c, where the approximation from Eq. ( 2 ) is only a small part of the picture. To recompute this experiment and find our new proposals, we model the SPDC sources as full squeeze operators in the Fock basis. We compute a (truncated) infinite series of amplitudes for arbitrary emission and absorption events 14 ; 20 . The two-mode SPDC is described as the following operator (detailed amplitudes in App. .1 ) S_{2}(\zeta)\left|p,q\right\rangle=\sum_{k=0}^{\infty}\sum_{n=0}^{\min{(p,q)}}\!\!\mathcal{A}^{(2)}_{k,n}(p,q;\zeta)\left|p\!-\!n\!+\!k,q\!-\!n\!+\!k\right\rangle, (3) where \zeta=re^{i\theta} , and for the single-mode case, S_{1}(\zeta)\left|p\right\rangle=\sum_{k=0}^{\infty}\sum_{n=0}^{\lfloor p/2\rfloor}\!\mathcal{A}^{(1)}_{k,n}(p;\zeta)\left|p\!+\!2(k\!-\!n)\right\rangle. (4) Figure 2: Optimization strategy. The search takes four steps: 1) experiment optimization for random sortings and target fidelities f_{0} (only a few shown), 2) simplification of the best candidates computed with a higher photon cutoff (convergence check), 3) optimization of previous solutions slowly varying f_{0} , and 4) final choice of best-performing designs. Figure 3: High probability entanglement by path identity. The presented experiments use SPDC crystals (squeeze operators) to generate entanglement. These path identity experiments are compared with previous designs, originally proposed for the low-gain regime, that neglected higher-order terms 18 ; 46 ; 45 . a) The first design produces a postselected W state of 4 qubits using threshold detectors. b) The second produces a heralded Bell state using 4 ancillary paths, each with a threshold detector. The entangled photons are not detected. c) The last one can generate two variants of N00N states using different parameters and postselection rules. For N=3, the threshold detector is not needed, i.e. \tau=0 , and the extra photons are lost. For N=4, we use a threshold detector and a single-mode squeezer acting on the ancilla mode. Automated experimental design — We want to produce entangled states with high success probability