Introduction

The diverse quantum phenomena and potential applications uniquely offered by ultracold molecules are owed to their intrinsic electric dipole moment and intricate internal structures. The promise of new science with molecules has driven significant advances in direct laser cooling and magneto-optical trapping of molecules in the past decade. Applications range from quantum simulation1,2,3,4,5,6 and computation7,8,9,10,11, to precision measurement tests for fundamental physics12,13,14,15,16,17,18, and studies of quantum-state-controlled chemistry and collisions19,20,21,22. The feasibility and further improvement of many of the current and proposed studies of cold molecules hinge on having a high number of trapped ultracold molecules. For example, in shot-noise limited precision measurements, such as those probing the electron electric dipole moment, the statistical uncertainty decreases with increasing number of trapped molecules N as \(\sim 1/\sqrt{N}\). With smaller and denser molecular clouds, loading molecules into an optical dipole trap becomes more efficient, thereby increasing N and sensitivity for experiments15. Similarly, to perform evaporative cooling towards the creation of quantum degenerate gases of ultracold molecules, a high initial density is necessary to facilitate rapid elastic collisions to occur, as well as to retain a sufficient number of molecules during the cooling process23,24,25. Finally, high densities are also desired in ultracold molecular collisional studies25.

The magneto-optical trap (MOT) is a pivotal technique for generating cold, dense samples of atomic gases, and has now been progressively extended to diatomic26,27,28,29,30,31 and polyatomic molecular gases32,33. It serves as an essential starting point for further cooling and subsequent transfer into conservative optical potentials34,35,36,37,38. Molecular MOTs are conventionally formed using red-detuned light ("red MOT"), where the trapping light is tuned below the transition resonance. These red MOTs typically contain 104 − 105 molecules, with a peak number density of  ~ 105 − 107 cm−3 29,30,39,40,41,42. These values are roughly five orders of magnitude lower than those routinely reached in atomic MOTs. Laser cooling of alkali atoms is typically realized on type-I transitions with \({F}^{\prime} > F\), where \(F({F}^{\prime} )\) is the quantum number of total angular momentum of the ground (excited) electronic state. In contrast, all laser cooling of molecules to date operates on type-II transitions (\({F}^{\prime} \le F\)) to ensure rotational closure for effective optical cycling30,43,44,45. Studies43 have shown that type-II red MOTs experience sub-Doppler heating at low velocities, producing a higher temperature and lower density cloud that has a limited overlap volume with the tightly-confined optical dipole trap (ODT). This, combined with molecular laser slowing limitations, lower photon scattering rate, and lower capture velocity of molecular MOTs compared to atomic MOTs in general, results in only a few thousand of trapped molecules in the optical dipole trap (ODT) and a phase-space density prohibitively low to realize many potential applications34,36,46. However, numerical simulations47 have revealed that for a type-II system in a 3D molasses, the sub-Doppler Sisyphus-type force near zero velocity persists over the range of magnetic fields typically seen in the MOT. This finding motivated the introduction of a blue-detuned MOT ("blue MOT") stage, utilizing sub-Doppler processes in the presence of a magnetic field and implemented after the red MOT to increase compression and, consequently, phase-space density. This approach, first demonstrated with Rb atoms47, has since been successfully implemented on several laser-cooled diatomic molecules, consistently producing substantially colder and denser samples37,48,49 with peak number densities of no ≈ 107 − 108 cm−3 and phase-space density of ρ ≈ 10−9. The reported cloud mean radii were ≳ 150 μm, limited by position-dependent trapping forces. These limitations prompted the recent development of a blue MOT50,51 using a frequency scheme that produces an additional physical effect, generating significantly more effective compression than the previously demonstrated blue MOTs. The improvement in the trapping force is attributed to Sisyphus-type blue-detuned cooling in the presence of a magnetic field gradient occurring in combination with a slowly-moving optical molasses50,51. This “conveyor-belt" effect shuttles the molecules towards the center region of the MOT and deposits them there.

Results

In this work, we experimentally realize a conveyor-belt blue-detuned MOT (simply, a conveyor MOT) for calcium monofluoride (CaF) molecules, capturing NMOT ≃ 1.3 × 105 molecules in a mean spatial radius of 64(5) μm and reaching a peak number density of \({n}_{o}^{\,{{\mathrm{MOT}}}} \,=3.6(5)\times 1{0}^{10}\,{{{{\rm{cm}}}}}^{-3}\), which is a more than 600-fold increase over the compressed red-detuned MOT, and the densest molecular MOT observed to date. We subsequently load the molecules into an optical dipole trap, trapping up to NODT ≃ 2.6 × 104 molecules at temperatures of TODT = 14(2) μK, achieving a peak number density \({n}_{o}^{\,{{\mathrm{ODT}}}}\,=1.05\times 1{0}^{10}\) cm−3 and a phase-space density as high as ρODT = 2.4 × 10−6. This NODT represents a 170-fold increase over the previously reported number for CaF molecules34 and, to the best of our knowledge, is the highest number of directly laser-cooled molecules in a conservative optical potential reported thus far. The molecule number density achieved here is comparable to those typically obtained in atomic optical dipole traps, providing a promising starting point of numerous near-term applications that rely on high densities of optically trapped ultracold molecules. The enhanced conveyor-belt trapping mechanism was first observed for CaOH molecules50, where the largely unresolved hyperfine structure renders its level scheme to resemble a two-level system. This raised the question of whether the approach would generalize to species with distinctively different hyperfine structure. Our work with CaF provides direct validation of the theoretical developments in ref. 51 by experimentally demonstrating that the mechanism extends to molecules with distinct and richer hyperfine structures.

The conveyor MOT mechanism

The underlying working principle of the conveyor MOT, illustrated in Fig. 1a and supported by numerical simulations51, is as follows.

Fig. 1: Working principle of a CaF conveyor MOT.
Fig. 1: Working principle of a CaF conveyor MOT.The alternative text for this image may have been generated using AI.
Full size image

a Illustration of the working principle of a conveyor-belt blue-detuned MOT; see text. The energy scale is shown for illustration purposes only. b Relevant level structure of CaF and laser configurations used in this work. J and F label the hyperfine states, where J is the total angular momentum without nuclear spin and F is the total angular momentum.

Consider a MOT beam containing two closely-spaced frequency components, ωa and ωb, each with opposite circular polarizations (σ+ and σ−, respectively). As in the conventional MOT configuration, the beams are directed towards the MOT center along, say, the \(\hat{x}\) direction, and retro-reflected, with the retro-reflected beam containing the same frequency components but flipped polarizations. When two counter-propagating beams of the same polarization interfere, they form a standing wave. If these beams have a relative detuning δ, the resultant wave will move along \(\hat{x}\) in the lab frame, creating the conveyor belt effect, much like a moving lattice. Here, the ωa (σ+) light traveling towards the \(+\hat{x}\) and the ωb (σ+) towards \(-\hat{x}\) will form a “σ+ conveyor belt" that slowly travels along \(-\hat{x}\) direction at a velocity \({v}_{{{{\rm{CB}}}}}\approx \frac{\delta }{2{\omega }_{a}}c\) where δ = ωb − ωa. Similarly, a “σ− conveyor belt" moving in \(+\hat{x}\) is formed by ωa (σ−) traveling towards \(-\hat{x}\) and ωb (σ−) in \(+\hat{x}\).

Now consider a type-II transition between hyperfine levels \(F=1\leftrightarrow {F}^{\prime}=0\), where the ground-state has a nonzero g-factor gF (assume, without loss of generality, that gF > 0). The presence of magnetic field gradients creates a position-dependent interaction strength between the molecule with either the σ+ or σ− conveyor belt, depending on the sign of gF. If ωa and ωb in the MOT beam are blue-detuned from the transition, in the B > 0 region, the \(\left\vert F=1,{m}_{F}=-1\right\rangle\) magnetic substate is Zeeman-shifted closer to resonance, and thus molecules preferentially interact with the σ+ conveyor belt. In the conveyor belt’s moving frame, the σ+ light appears to the molecules as a stationary wave, setting up a spatially periodic light shift potential to facilitate Sisyphus-type cooling43. Here, a molecule experiences maximum optical pumping rate as it traverses across the intensity maxima, before spontaneously decaying to the \(\left\vert F=1,{m}_{F}=0,1\right\rangle\) dark states. The molecule subsequently returns to the \(\left\vert F=1,{m}_{F}=-1\right\rangle\) bright state primarily through orthogonal MOT beams and the transverse magnetic field, thereby completing the Sisyphus cooling cycle. In the lab frame, the molecules get accelerated towards  ~ vCB and transported towards the MOT center by the moving σ+ conveyor belt. On the other side of the MOT, in the B < 0 region, the molecules preferentially interact with the σ− conveyor belt while experiencing Sisyphus cooling in the σ− conveyor belt’s moving frame. Near the MOT center, where the magnetic field is weak, the preferential interaction with the σ± conveyor belts diminishes. At the center, the dominant cooling is provided by gray-molasses cooling in the lab frame with Λ-systems forming between the hyperfine levels. This, in part, effectively prevents the molecules from being over-transported out of the MOT center region. In summary, in a conveyor MOT, molecules away from the MOT center on either sides selectively interact with one of the σ± conveyor belts, in which they are cooled into the moving frame of the conveyor belt and slowed to near zero velocity as they approach the center, resulting in a large number of near zero-velocity molecules piled up in the MOT center.

Capturing molecules into a conveyor MOT

In our experiment, CaF molecules are initially captured in a radio-frequency (RF) red MOT operating on the \(X(\nu=0,N=1)-A({\nu}^{\prime}=0,J{\prime}=1/2)\) cooling transition29. The apparatus and optical cycling scheme are described in our previous work29,52,53. We begin with a magnetic field gradient of \({B}^{{\prime} }=\) 20 G/cm RMS and a total intensity of Io = 30 mW/cm2 per MOT beam at a detuning of Δ = 2π × − 8 MHz and load the MOT for 5 ms. The intensity is then reduced to Io/4 in 5 ms, and the magnetic field gradient is ramped to \(B{\prime}=\) 50 G/cm RMS, compressing the molecular cloud to σ ~ 600 μm, with 2 × 105 trapped molecules, corresponding to \({n}_{o}^{\,{{\mathrm{red}}}}\,\approx \,5.8\times 1{0}^{7}{{{\mathrm{cm}}}}^{-3}\). In the next 5 ms, we turn off the RF magnetic field and polarization-switching, and jump the laser intensity and detuning to the free-space Λ-cooling optical scheme35,54. A subsequent 5-ms sub-Doppler cooling pulse reduces the temperature of the molecular cloud to 18(5) μK while retaining  ~ 80% of the initially captured molecules.

Next, we load the molecules from the red MOT into a DC conveyor MOT. To do so, we first switch the MOT light frequency to the configuration shown in Fig. 1b by introducing one additional frequency component to form the conveyor MOT. In this configuration, the frequency ωI, addressing the \(\left\vert F=1-\right\rangle\) hyperfine state, is blue-detuned with a single-photon detuning Δ. The other two frequency components, ωa and ωb, both addressing the \(\left\vert F=2\right\rangle\) state, have opposite circular polarizations and detunings of Δ + δa and Δ + δb, respectively. These two closely-spaced components with a slight relative detuning of δ = δb − δa create the conveyor-belt effect and consequently an enhanced trapping force on the \(\left\vert F=2\right\rangle\) state, while ωI mainly functions as a repumper and facilitates lab-frame cooling at the MOT center. We then gradually increase the DC magnetic field using an optimized ramp profile that minimizes molecular loss. We realize the switch from RF to DC MOT by employing high-impedance inductors and back-to-back solid-state relays (See “S2. RF - DC MOT Coil Switching" section in the Supplementary Information for more details). Finally, to detect and measure the spatial profile of the molecular cloud, we collect in-situ fluorescence onto a camera.

Dependence of molecular cloud size on frequency detunings and magnetic field gradient

To maximize compression of the molecular cloud, we investigate the effect of single and two-photon detunings (Δ, δa and δb) on the mean radius σ of molecules in the conveyor MOT. Figure 2a–c show σ as a function of Δ, δa and δb. We obtain σ in each direction by fitting it to a Gaussian distribution. σ is minimized at δa = − 2π × 1.5 MHz, δb = 2π × 0.5 MHz, corresponding to a conveyor belt speed of approximately vCB ≈ 0.6 m/s. Physically, due to the decreasing magnetic field as the molecules are compressed towards the MOT center, combined with the frictional force imparted by the opposing conveyor belt, the molecules on average do not get fully accelerated to vCB. σ is relatively insensitive to Δ over the range Δ = 2π × (20 MHz to 30 MHz), similar to the observations in ref. 50. This detuning range suggests that part of the trapping force may also originate from the \(\left\vert F=1+\right\rangle\) state. Δ = 2π × 20 MHz is used for all subsequent measurements. In Fig. 2d, a scan of \({B}^{{\prime} }\) reveals that σ = 64(5) μm (with radial and axial widths of σax = 58(4) μm and σrad = 68(2) μm, respectively) is reached at \({B}^{{\prime} }\approx 40\,\,{{\mathrm{G/cm}}}\,\). The light intensity used for the conveyor MOT is  ~ 2Isat and the total light power is equally distributed among the three frequency components. We observe that the minimum σ is relatively insensitive to the ratio of power intensities for the three components. With the available laser power in our experiment, we observe an increase in the number of trapped molecules with higher power (Supplementary Fig. S4), as expected from the corresponding enhancement in both capture velocity and trap volume for gray-molasses cooling.

Fig. 2: Dependence of conveyor MOT cloud size on frequency detunings and magnetic field.
Fig. 2: Dependence of conveyor MOT cloud size on frequency detunings and magnetic field.The alternative text for this image may have been generated using AI.
Full size image

a MOT radius σ as a function of two-photon detunings (a) δa, (b) δb, (c) single-photon detuning Δ, and (d) applied magnetic field gradient \({B}^{{\prime} }\). For each scan in (a–c), all other detunings and the total intensity are kept fixed. The intensity of each beam is calibrated such that it remains constant as frequency is varied. The maximum magnetic field gradient used in (a–c) is \({B}^{{\prime} }\approx 40\) G/cm.

Figure 3 presents σ and no as a function of conveyor MOT application time, with top images showing the in-situ fluorescence. Most of the compression takes place within the first 10 ms, during which the cloud reduces from σ ≳ 550 μm to σ = 130μm as \({B}^{{\prime} }\) is ramped to  ~ 20 G/cm. Over the next 20 ms, σ continues to decrease as \({B}^{{\prime} }\) reaches 40 G/cm. By the end of the compression, up to NMOT ~ 1.3 × 105 molecules remain trapped, corresponding to a peak density of \({n}_{o}^{\,{{\mathrm{MOT}}}} \,=\,3.6(5)\times 1{0}^{10}\) cm−3, with measured mean temperature of TMOT = 141(7) μK. Given the relatively low measured photon scattering rate of 0.27(3) × 106s−1 in the conveyor MOT (using laser parameters optimized for minimal σ) (Supplementary Fig. S2), it may be that higher light intensity could provide a larger trapping volume, but this is open for further investigation. In addition, we find that the 1/e lifetime of molecules in the conveyor MOT follows a single exponential decay, with a fitted τ = 37.2(8) ms measured for \({B}^{{\prime} }\approx 40\) G/cm (Supplementary Fig. 1), limited by the trap loss at high magnetic field due to the lowered trap depth.

Fig. 3: Size and density of conveyor-belt blue MOT as a function of time.
Fig. 3: Size and density of conveyor-belt blue MOT as a function of time.The alternative text for this image may have been generated using AI.
Full size image

Blue: fitted 1/e cloud radius σ (left axis). Orange: peak molecule number density no (right axis). Dashed lines are guides to the eye. Data are taken with frequency parameters Δ = 2π × 20 MHz, δa = − 2π × 1.5 MHz and δb = 2π × 0.5 MHz, and a magnetic field gradient of up to \({B}^{{\prime} }\approx 40\) G/cm. Top: Images of the molecular cloud taken by collecting 2 ms of in-situ fluorescence following a compression time of t.

Loading molecules from a conveyor MOT into an ODT

The high molecular density obtained in the conveyor MOT provides a reservoir from which we transfer molecules into a 1064 nm ODT of 57 μm beam waist and a trap depth of U ≈ kB × 174 μK. To load the ODT, we first switch off all MOT light and DC magnetic field, then immediately turn on the ODT light in the presence of Λ-cooling light for 3 ms. This sequence prevents the ODT trap light from inducing significant differential Stark shifts that would influence the conveyor MOT operating parameters. The number of molecules transferred into the ODT depends on σ of the cloud reservoir, as expected, as shown in Fig. 4a. The loading efficiency increases as the size of the conveyor MOT cloud approaches that of the ODT, leading to better spatial mode-matching between the two. Up to 24% of the molecules from the cloud reservoir are captured into the optical trap, resulting in NODT ≃ 2.6 × 104 molecules at TODT = 14(2) μK with ρODT = 2.4 × 10−6. This corresponds to 12% of the molecules initially trapped in the red MOT ending up in the ODT, a factor of five improvement over our previous loading efficiency of ≲ 2% using direct transfer from the red MOT.

Fig. 4: ODT loading.
Fig. 4: ODT loading.The alternative text for this image may have been generated using AI.
Full size image

a Loading efficiency as a function of conveyor MOT cloud size. b Number of molecules in the ODT as a function of loading time for different initial numbers of molecules. c Fraction of molecules remaining in the ODT as a function of hold time in the presence of Λ-cooling light, fitted to a one-body (red, dashed) and two-body (blue, dashed) loss models. d Expansion of the reservoir cloud in the presence of Λ-cooling as a function of time. The ballistic expansion trajectory for cloud at 147 μK is plotted as a guide (red, dashed). A diffusive model fitted to the data is shown (green, dashed).

To understand the possible loss mechanisms during the loading process, we examine the time dependence of the molecular number during the ODT loading. Figure 4b shows the loading curves of NODT at varying NMOT in the cloud reservoir, NMOT. The ODT loading dynamics are well described by a rate equation model55 by accounting for overall molecular loss in the reservoir cloud, one-body, and two-body loss mechanisms (see “S5. ODT Loading Dynamics and Losses" section in Supplementary Information). To investigate the potential role of two-body collisional loss in limiting the loading of molecules in the ODT, we measure the lifetime of molecules in the ODT in the presence of varying Λ-light duration21,56. Fitting the data to both one-body and two-body loss models yields decay rates that are statistically indistinguishable (Fig. 4c). In addition, the linear scaling observed between NODT and NMOT suggests that current loading is far from saturation.

The results in Fig. 4d reveal that the effective ODT loading time, as seen in Fig. 4b, is limited by the rapid expansion of the molecular cloud. σ increases to 153 μm in the first 3 ms in the presence of Λ-cooling light, leading to a large spatial mismatch with the ODT. With a typical Λ-cooling time of  < 300μs, the expansion quickly deviates from a ballistic trajectory that would be expected for the 147 μK conveyor MOT temperature, instead following a diffusion trajectory (dashed lines in Fig. 4(d)). This rapid diffusive expansion is the primary limitation on our loading efficiency, and larger optical dipole trap volume would likely improve it. In the future, the cloud expansion issue might be addressed by directly loading the ODT from the conveyor MOT, provided the optical trap depth is much higher than the conveyor MOT cloud temperature.

Discussion

In summary, we realize a high-density MOT of CaF using the conveyor-belt mechanism, which provides substantially higher trapping force while operating at a relatively lower scattering rate compared to a conventional blue MOT of CaF49, and demonstrate its applicability beyond a simpler, previously studied system50. Together with50 and51, these results establish a versatile and broadly applicable method for producing dense samples of laser-cooled molecules. The strong trapping forces of the conveyor MOT further enable the creation of highly compressed molecular samples, which subsequently allows efficient transfer into an optical dipole trap within the expansion time of the reservoir cloud. As a result, approximately 2.6 × 104 molecules are loaded into an optical dipole trap at a phase-space density of 2.4 × 10−6. The high molecular number and phase-space density in optical traps present immediate possibilities for a range of near-term applications, including enhanced precision measurement with ultracold molecules, improved loading into optical tweezer arrays, and studies of two-body collisions in bulk molecular gases22. Moreover, these improvements bring evaporative cooling within reach, paying the way towards a quantum degenerate gas of laser-cooled molecules. The conveyor-belt effect demonstrated here holds robustly within the studied parameter space, with the possibility of more complex dynamics emerging in other regimes51. Owing to its simplicity and effectiveness, the conveyor MOT offers a broadly applicable approach for molecular MOTs and type-II atomic systems.

Methods

Overview of experimental sequence

The experiment proceeds through four general stages:

  1. (i)

    RF MOT. CaF molecules are initially captured in a radio-frequency (RF) red-detuned MOT operating on \(X(\nu=0,N=1)-A({\nu}^{\prime}=0,J{\prime}=1/2)\) cooling transition at 606 nm, as described in refs. 29,52, with three pairs of orthogonal beams, each addressing all four ground-state hyperfine components in the N = 1 rotational manifold. Frequency sidebands generated by acousto-optic modulators enable independent control of polarization and intensity, while rapid polarization switching is provided by a Pockels cell. A 1-ms resonant light pulse is used to image the molecules, which is then converted to a molecular number following a calibration procedure described in the appendix of our previous work52.

  2. (ii)

    Sub-Doppler cooling and Λ cooling. In the next 5 ms, the RF field and Pockels cell are turned off, and the MOT light parameters are jumped to those required for sub-Doppler cooling35, single-photon detuning of Δ = + 34 MHz. A subsequent 5-ms Λ-cooling pulse reduces the cloud temperature to a base temperature of 18(5) μK while retaining  ~ 80% of the loaded molecules.

  3. (iii)

    Conveyor MOT compression. We transfer the molecules into a DC conveyor MOT by switching the MOT light frequency to the scheme shown in Fig. 1b and converting the magnetic field from RF to a DC configuration. Over 30 ms the cloud compresses to 64(5) μm, after which we hold the magnetic field for an additional 10 ms to collect the in-situ fluorescence signal to obtain the spatial profile of the compressed cloud.

  4. (iv)

    Optical dipole trap (ODT) loading. After switching off all MOT lights and the DC magnetic field, we turn on the 1064 nm trap light with Λ-cooling light applied for 3 ms. The molecules are held in the ODT for 100 ms to allow the background cloud to fully dissipate, and then probed with a 1-ms resonant pulse. To obtain stable ODT loading efficiency measurements, for each experimental cycle, we use a 5 ms in-situ fluorescence collected at the end of the conveyor MOT compression as a reference signal to remove fluctuations in the molecular flux from the cryogenic buffer gas source.

Additional details can be found in Supplementary Information.

Measurements of conveyor MOT lifetime and scattering rate

The conveyor MOT lifetime after the compression stage is determined by holding the compressed cloud at a fixed magnetic field gradient of 40 G/cm and monitoring the fluorescence decay over time. We obtain a single-exponential lifetime of τconveyor= 37(8) ms, consistent with the absence of light-assisted two-body loss at this density. To determine the scattering rate, selected vibrational repump lasers were turned off, limiting the photon budget per molecule. From the resulting fluorescence decay, we extract a scattering rate of 0.27(3) × 106s−1 (Supplementary Fig. 1).

Measurements of cloud temperature

The temperature of the molecular cloud is determined using standard time-of-flight expansion, assuming a Gaussian density distribution in the MOT and ODT. After being released from the trap, the molecular cloud is allowed to freely expand for variable durations before being resonantly imaged for ; the 1 ms exposure time here is chosen as it is short compared to the expansion time and causes negligible heating. Gaussian fits to the radial and axial profiles yield Trad and Tax, and the reported temperature is given by the geometric mean \(T={({T}_{{{{\rm{rad}}}}}^{2}{T}_{{{{\rm{ax}}}}})}^{1/3}\).