betaFeedback

WorksPhysics and Astronomy

Energy- and momentum-resolved Cooper-pair emission from a cuprate superconductor

Abstract

Superconductivity arises from Cooper pairing, but conventional photoemission probes electrons individually rather than as pairs. Here we report the observation of Cooper-pair emission from an optimally doped cuprate superconductor using two-electron angle-resolved photoemission (2e-ARPES). Below the superconducting transition, we observe a distinct two-electron emission peak at zero pair binding energy and zero pair momentum that disappears in the normal state, revealing the characteristic fingerprint of a Cooper pair. Unexpectedly, the joint energy distribution broadens markedly along the relative-energy axis as temperature decreases, while the total energy width remains nearly unchanged. This anomalous behavior cannot be explained by instrumental resolution, static disorder or thermal broadening, and instead points to scattering processes involving collective excitations of the superconductor. Our results establish 2e-ARPES as a probe of both Cooper-pair emission and the many-body dynamics of the superconducting state.

Superconductivity arises when electrons form Cooper pairs and develop long-range phase coherence [1, 2]. Yet, most spectroscopic probes of superconductors access only single-particle excitations. Angle-resolved photoemission spectroscopy (ARPES), scanning tunneling spectroscopy, transport, and optical measurements have established the phenomenology of several superconductors, revealing the gap structure, quasiparticle excitations, and intertwined orders [3, 4, 5, 6, 7, 8]. However, these techniques probe superconductivity only indirectly – they measure how individual electrons are affected by pairing, rather than directly accessing the two electrons that form a Cooper pair.

That limitation is especially acute in the cuprates, in which high-TcT_c superconductivity emerges from a strongly interacting electronic environment characterized by short coherence lengths, pronounced momentum-space anisotropy, and the presence of competing orders, all developing from a normal state that departs strongly from conventional Fermi-liquid behavior [4, 5, 7, 8, 9]. In such materials, signatures observed in the single-particle spectrum do not by themselves reveal whether the underlying paired state is coherent, nor do they determine the pair momentum. A direct probe of pairing should therefore both establish whether the initial state supplies a coherent electron pair and determine the pair momentum, K=k1+k2\mathbf{K} = \mathbf{k}_1 + \mathbf{k}_2, where ki\mathbf{k}_i is the measured in-plane momentum of electron ii. For a spatially uniform, phase-coherent superconducting condensate, such a measurement should reveal a pair signal concentrated near K=0\mathbf{K} = 0, whereas weight at finite K\mathbf{K} could indicate finite-momentum pairing. Furthermore, resolving the energies of both electrons could provide access to the interactions and excitations of the material that influence the pair-emission process.

Superconductivity arises when electrons form Cooper pairs and develop long-range phase coherence [1, 2]. Yet, most spectroscopic probes of superconductors access only single-particle excitations. Angle-resolved photoemission spectroscopy (ARPES), scanning tunneling spectroscopy, transport, and optical measurements have established the phenomenology of several superconductors, revealing the gap structure, quasiparticle excitations, and intertwined orders [3, 4, 5, 6, 7, 8]. However, these techniques probe superconductivity only indirectly – they measure how individual electrons are affected by pairing, rather than directly accessing the two electrons that form a Cooper pair.

That limitation is especially acute in the cuprates, in which high-TcT_c superconductivity emerges from a strongly interacting electronic environment characterized by short coherence lengths, pronounced momentum-space anisotropy, and the presence of competing orders, all developing from a normal state that departs strongly from conventional Fermi-liquid behavior [4, 5, 7, 8, 9]. In such materials, signatures observed in the single-particle spectrum do not by themselves reveal whether the underlying paired state is coherent, nor do they determine the pair momentum. A direct probe of pairing should therefore both establish whether the initial state supplies a coherent electron pair and determine the pair momentum, K=k1+k2\mathbf{K} = \mathbf{k}_1 + \mathbf{k}_2, where ki\mathbf{k}_i is the measured in-plane momentum of electron ii. For a spatially uniform, phase-coherent superconducting condensate, such a measurement should reveal a pair signal concentrated near K=0\mathbf{K} = 0, whereas weight at finite K\mathbf{K} could indicate finite-momentum pairing. Furthermore, resolving the energies of both electrons could provide access to the interactions and excitations of the material that influence the pair-emission process. [10]. A promising route to such a probe is two-electron angle-resolved photoemission spectroscopy (2e-ARPES). In this approach, absorption of a high-energy photon ejects two photoelectrons whose energies, E1E_1 and E2E_2, and in-plane momenta, k1\mathbf{k}_1 and k2\mathbf{k}_2, are recorded in coincidence (Figure 1a). Here, EiE_i is measured relative to the Fermi level, and the coincidence events define the full energy- and momentum-resolved distribution P(2)(E1,k1;E2,k2)P^{(2)}(E_1,\mathbf{k}_1;E_2,\mathbf{k}_2). This approach builds on more than four decades of attempts at electron-coincidence double photoemission. Beginning with Auger measurements in Cu [11], one-photon double photoemission from valence states was subsequently demonstrated in Cu and Ni and extended to insulators and correlated oxides [11, 12] [13, 14, 15, 16, 17, 18]. Theoretical work predicts that a 2e-ARPES measurement of superconductors contains a contribution arising directly from the superconducting condensate and sensitive to the momentum and internal structure of Cooper pairs [19–23]. For electrons emitted from paired states near the Fermi surface, the simulated condensate contribution produces a pronounced peak centered at E1=E2=0E_1 = E_2 = 0 in the joint energy distribution, as illustrated in Figure 1b (see Supplementary Information for simulation details). Yet, despite longstanding interest, this condensate peak associated with superconductivity has not previously been observed. The challenge is to combine a sufficiently high photoemission coincidence rate with energy resolution on the superconducting scale and momentum-selective detection near opposite points of the Fermi surface [17].

Figure 1. Two-electron ARPES geometry (a) Schematic of the two-electron angle-resolved photoemission spectroscopy (2e-ARPES) process, in which absorption of a single XUV photon leads to the coincident emission of two photoelectrons that are detected simultaneously. (b) Simulated 2e-ARPES joint energy distribution for Cooper-pair emission from a dd-wave superconductor with zero pair momentum (k1=−k2\mathbf{k}_1 = -\mathbf{k}_2; see lower left inset showing opposite antinodal momenta on the typical cuprate Fermi surface). Here, P(2)P^{(2)} is the normalized pair-photoemission intensity (a.u.), and E1E_1 and E2E_2 are the photoelectron energies relative to the Fermi level. The condensate contribution peaks at E1=E2=0E_1 = E_2 = 0. Simulation details are provided in the Supplementary Information.

In this work, we perform 2e-ARPES measurements on the optimally-doped cuprate superconductor Bi2Sr2CaCu2O8+δ\mathrm{Bi}_2\mathrm{Sr}_2\mathrm{CaCu}_2\mathrm{O}_{8+\delta} (Bi-2212) (Tc=93T_c = 93 K), focusing on photoelectron pairs near the antinodal regions of the Brillouin zone. Our purpose-built 2e-ARPES setup combines a high-flux, 525-kHz XUV source with better than 15-meV energy resolution, overcoming the instrumental limitations that have previously prevented the superconducting condensate peak from being resolved (Methods and [24]). In the joint energy map P(2)(E1,E2)P^{(2)}(E_1,E_2), we observe a zero-pair-binding-energy peak that appears below TcT_c and is absent above it, providing direct evidence for the emission of correlated electron pairs from the superconducting state. The same events are concentrated near zero pair momentum, K=0\mathbf{K} = 0. To our knowledge, this is the first momentum- and energy-resolved observation of Cooper-pair emission from a superconductor. We further find that the widths along E+E_+ and E−E_- evolve differently. The total-energy width is nearly temperature independent, whereas the relative-energy width broadens strongly on cooling. We argue that this behavior provides insights into the correlations present inside the superconducting state.

We first characterize the electronic structure of our Bi-2212 samples using conventional ARPES. In this configuration, the sample normal is oriented at 45° relative to each analyzer, enabling simultaneous detection of photoelectrons from opposite momentum regions of the Brillouin zone. For our photon energy of 22.8 eV, the expected momentum regions probed by conventional ARPES in this geometry are shown by the blue shaded regions 1 and 3 in Figure 2a (See Supplementary Information for details and calibration calculations). The resulting constant energy cuts near the Fermi level for each detector are shown in Figure 2b. To further characterize the sample, we also measure the ARPES spectra around the center of the Brillouin zone (region 2 in Figure 2a) by rotating the sample such that its surface normal is oriented along the central axis of the analyzer. Overall, the measured spectra reproduces the expected Fermi-surface of optimally doped Bi-2212 in each of the probing regions [25].

Figure 2. Single-particle ARPES characterization and calibration. (a) Schematic of the momentum regions probed by single-ARPES (blue shaded regions numbered 1 and 3) and 2e-ARPES (red shaded regions 4 and 5) at a photon energy of 22.8 eV in the experimental geometry shown in Figure 1, where the sample normal is oriented at 45° relative to each analyzer. Region 2 is accessed by rotating the sample normal along the central axis of one analyzer (see Supplementary Information for calculations). (b) Constant-energy cuts near the Fermi level as measured at the center of the Brillouin zone and in regions 1 and 3 in (a). Data from regions 1 and 3 are acquired simultaneously using the two detectors, while region 2 is measured with the rotated sample geometry. An energy-and-momentum dispersive cut along the gray dotted line is shown in the top panel.

Establishing a reliable Fermi-level reference is essential for interpreting the two-electron coincidence measurements. We therefore exploit the well-known nodal structure of the Bi-2212 pairing state, where the superconducting gap vanishes along the nodal direction [7] [26]. This nodal point appears at the center of a Fermi arc and provides a robust, gapless reference for energy calibration (Supplementary Information). We determine the photoelectron kinetic energy corresponding to the Fermi level by analyzing the Fermi–Dirac cutoff across this nodal Fermi edge on both detectors. This calibration establishes the Fermi-level reference, relative to which E1E_1 and E2E_2 are defined for all subsequent 2e-ARPES measurements performed with the same sample orientation, detector geometry, and experimental conditions.

We next perform 2e-ARPES measurements using the same sample orientation and detector geometry as in the conventional ARPES measurements. The setup is configured to detect correlated pairs of photoelectrons emitted near the Fermi energy (around zero binding energy). In contrast to single-electron photoemission, the correlated pair of electrons generated by absorption of a single photon share the available kinetic energy, and therefore arrive at the detectors with substantially lower kinetic energies than electrons emitted in conventional ARPES from the same binding energy [17, 27]. As a consequence, the momentum-space region accessed in 2e-ARPES is shifted relative to the corresponding conventional ARPES measurement. The expected momentum regions probed in the two-electron measurement are shown schematically in Figure 2a by the red shaded regions 4 and 5. These regions contain the antinodal points of the Fermi-surface of Bi-2212, thus enabling direct access to the maximum superconducting gap [21].

Figure 3 shows the 2e-ARPES data acquired around the Fermi level. Measurements were performed at two temperatures: 8.5 K, well below the superconducting transition temperature Tc=93T_c = 93 K, and 200 K, far above TcT_c. The data are analyzed using the normalized correlation measure P(2)P^{(2)}, which quantifies the relative intensity of true coincidence events after subtraction of the uncorrelated background (Methods and Supplementary Information). The full two-electron dataset is intrinsically six-dimensional, with each coincidence event characterized by the energy and two in-plane momentum components of each electron. To visualize the correlated signal, we first integrate over momentum and plot the joint energy map P(2)(E1,E2)P^{(2)}(E_1,E_2). These joint energy maps provide an overview of how the absorbed photon energy is distributed between correlated photoelectrons.

Figure 3. 2e-ARPES measurements on Bi-2212 below and above TcT_c (a,b) Joint energy maps of the normalized correlation measure P(2)(E1,E2)P^{(2)}(E_1,E_2), integrated over the momentum regions shown in Figure 2a, measured at 8.5 K and 200 K, respectively. Here, E1E_1 and E2E_2 are the photoelectron energies relative to the Fermi level. The solid and dashed arrows in (a) indicate the diagonal (E+=E1+E2E_+ = E_1 + E_2) and anti-diagonal (E−=E1−E2E_- = E_1 - E_2) directions, respectively. (c) Cuts along the diagonal (E+E_+) direction in (a) and (b). Cuts along the anti-diagonal (E−E_-) direction are shown in Figure 4. Shaded regions around the curves represent the 68% confidence interval calculated from bootstrap resampling (see Methods and Supplementary Information). (d) Pair-momentum distribution constructed from coincidence events within an approximately 75 meV × 75 meV energy window centered at E1=E2=0E_1 = E_2 = 0 in (a).

Figure 3a reveals a striking feature at 8.5 K, with a sharp, well-defined peak in the two-electron energy distribution. The peak is centered at E1=E2=0E_1 = E_2 = 0, corresponding to zero pair-binding energy as established by conventional ARPES, with intensity concentrated along the diagonal E−=E1−E2=0E_- = E_1 - E_2 = 0. In contrast, measurements performed at 200 K (Figure 3b) show a pronounced suppression of the correlated signal and no discernible peak is observed at zero binding energy. The presence (absence) of the zero-energy peak at 8.5 K (200 K) is also evident in the diagonal cuts of P(2)(E1,E2)P^{(2)}(E_1,E_2) shown in Figure 3c.

The presence of a zero-energy peak in P(2)(E1,E2)P^{(2)}(E_1,E_2) below TcT_c is a clear signature of Cooper-pair photoemission from the superconducting condensate and has been predicted to occur in previous theoretical work on double photoemission from superconductors [20, 21]. In particular, as first noted by [20], double photoemission from a superconducting state exhibits a qualitatively distinct spectral feature from single-particle photoemission: a peak at zero pair-binding energy that exists only below TcT_c. The physical origin of this distinct feature lies in the nature of the excitation created by the photon. In conventional photoemission, absorption of a photon leads to the ejection of an electron from a Cooper pair, which requires a binding energy of Δk\Delta_{\mathbf{k}}. In contrast, in double photoemission the photon removes both electrons from a Cooper pair simultaneously which does not break the Cooper pair and thus does not cost any binding energy, therefore producing a peak at zero energy.

To further confirm that the zero-energy peak originates from the superconducting condensate, we next examine the momentum structure of the correlated electron pairs by restricting the analysis to a narrow energy window centered around the origin in Figure 3a. To characterize the pair momentum, we construct a pair-momentum map of P(2)P^{(2)} using the same subset of coincidence events (Figure 3d). In this representation, the horizontal and vertical axes are the center-of-mass momentum components Kx=k1x+k2xK_x = k_{1x} + k_{2x} and Ky=k1y+k2yK_y = k_{1y} + k_{2y}. The resulting distribution thus provides a direct measure of the pair momentum. A pronounced maximum is observed at K=0\mathbf{K} = 0, indicating that the dominant contribution to the correlated signal arises from electron pairs with zero pair momentum, as expected for Cooper pairing in optimally doped Bi-2212 [20, 21]. At least for optimal doping, no signature of a pair-density wave, corresponding to a peak at non-zero K\mathbf{K} [21], is observed [28].

Having established the superconducting condensate-associated feature, we next examine its temperature dependence in the two-dimensional energy distribution. We consider two coordinates, the total energy E+=E1+E2E_+ = E_1 + E_2 and the relative energy E−=E1−E2E_- = E_1 - E_2, corresponding to the diagonal and anti-diagonal directions of the joint energy map, respectively. These coordinates distinguish changes in the total detected energy from changes in how that energy is shared between the two electrons.

Figures 4(a)-(c) show the evolution of the joint energy distribution at 8.5 K, 30 K and 70 K. The most striking feature is that the width along the relative-energy direction (anti-diagonal, Figure 4(e)) increases substantially upon cooling, whereas the width along the total-energy direction (diagonal, Figure 4(d)) remains nearly unchanged. Lorentzian fits to the experimental cuts yield a relative-energy full width at half maximum that increases from approximately 12 meV at 70 K to 23 meV at 8.5 K, as shown in Figure 4(f). Thus, as superconductivity strengthens upon cooling, the detected two-electron energy distribution becomes increasingly broadened in the relative-energy coordinate without a corresponding increase in its total-energy width.

Figure 4. Temperature dependence of diagonal and anti-diagonal widths in the 2e-ARPES signal. (a)-(c) Contour plots of P(2)(E1,E2)P^{(2)}(E_1,E_2) at 8.5 K, 30 K and 70 K. Similarly colored lines represent contours along the same relative intensity. Diagonal cuts (d) and anti-diagonal cuts (e) of P(2)(E1,E2)P^{(2)}(E_1,E_2) at 8.5 K, 30 K and 70 K. Solid lines are Lorentzian best fits to the data. (f) Full width at half maximum (FWHM) extracted from Lorentzian fits to the anti-diagonal cuts. Error bars indicate the 68% confidence interval of the Lorentzian fits from (e), calculated from the uncertainties in the fitting parameters.

This contrasting temperature dependence cannot be explained by the temperature-independent instrumental resolution nor by the thermal broadening commonly present in standard ARPES, as the latter should decrease as the temperature is lowered. Rather, it suggests that the double-photoemission process is subject to an energy redistribution that can only arise from its interaction with the collective degrees of freedom of the superconductor itself. This, in turn, would imply that the joint energy map P(2)(E1,E2)P^{(2)}(E_1,E_2) provides unique access to excitations present inside the superconducting state.

While determining the microscopic mechanism responsible for this enhanced anti-diagonal broadening is beyond the scope of our work, we discuss two complementary scenarios. In the first one (Supplementary Information Sec. IV), the first ejected photoelectron undergoes intermediate-state Coulomb scattering off electrons in the material before the second photoelectron is emitted. Such scattering processes transfer energy and momentum between the first photoelectron and the electronic system, thus naturally explaining the relative difference in two photo-electrons’ energies. Since the second photo-electron can be subject to the same scattering processes after emission, a finite distribution of detected photoelectron energies would result. We note that these scattering processes would necessarily lead to a coupling of the photo-electrons to collective spin or charge excitations inside the superconductor [29–31], thus directly probing its strongly correlated nature.

A second possibility involves the relative-time dynamics of the outgoing two-electron state. In this description (Supplementary Information Sec. V), the two high-energy electrons are initially generated with equal energy (E−=0E_{-} = 0) by an effective pair-emission process. However, each outgoing electron subsequently interacts with collective excitations present in the superconducting state of the material, resulting in different scattering histories. This introduces an additional relative-time dependence into the two-electron photoemission amplitude, broadening the measured distribution of the energy difference between the electrons. Within this framework, the relative-energy linewidth reflects both the lifetime or dephasing of the outgoing electrons due to their coupling to these collective excitations. A natural source is fluctuations of the superconducting order parameter [32–34], which provide a time-dependent and spatially non-uniform environment through which the two photoelectrons propagate. In the cuprates, the short coherence length, strong momentum dependence of the gap and electronic inhomogeneity [35–37] can make these fluctuations particularly effective at giving the two electrons different scattering histories. As the condensate amplitude strengthens on cooling, it can overcome the suppression of the thermal excitation of these modes, producing the larger anti-diagonal linewidth observed at low temperature.

Both mechanisms illustrate how the energy distribution in double photoemission can carry information beyond the properties of the Cooper pair itself. The observation of a superconducting condensate-associated peak establishes 2e-ARPES as an experimental probe of Cooper-pair emission, while its unexpected linewidth evolution points toward a broader opportunity to investigate the interactions and dynamics accompanying pair removal in correlated materials.

Acknowledgments

Acknowledgments: We thank Peter Abbamonte, Peter Armitage, Juan C. Campuzano, Tai-Chang Chiang, Eugene Demler, Vidya Madhavan, Mike Norman, Mohit Randeria, George Sawatzky, Z.-X. Shen, and Lucas Wagner for helpful discussions. The work by H.A., J.Z., and F.M. was supported by the EPiQS program of the Gordon and Betty Moore Foundation, Grant GBMF11069, and by the U. S. Department of Energy, Office of Science, Basic Energy Sciences, under Award No. DE-SC0021238 (project conception). The work at BNL was supported by the U. S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division, under contract no. DE-SC0012704. The work by K.H.W. and D.K.M was supported by the U. S. Department of Energy, Office of Science, Basic Energy Sciences, under Award No. DE-FG02-05ER46225. J.N. was partly supported by the U. S. Department of Energy, Office of Science, Office for Nuclear Physics under award no. DE-SC0023861. The authors acknowledge the use of facilities and instrumentation supported by NSF through the University of Illinois at Urbana-Champaign Materials Research Science and Engineering Center DMR-2309037. Sample characterization was carried out in part in the Materials Research Laboratory Central Research Facilities, University of Illinois.

Author contributions: H.A. and J.Z. performed the measurements and analyzed the experimental data. G.G. synthesized the optimally doped Bi-2212 single crystals. K.H.W. and D.K.M. performed the simulation presented in Figure 1b and in SI Sec.I. D.K.M. developed the theoretical description of intermediate-state Coulomb scattering during double photoemission (SI Sec. IV). R.M.F. and J.N. developed the theoretical framework relating relative-energy broadening to relative-time dynamics and dephasing of the emitted two-electron state (SI Sec. V). All authors contributed to the writing of the manuscript. F.M. supervised the project.

Competing interests: The authors declare no competing interests.

Methods

Experimental setup: Our 2e-ARPES setup uses a gas based high-harmonic generation (HHG) beamline as a photon source. A Yb:KGW laser (CARBIDE, Light Conversion) provides a 1030 nm fundamental beam with a pulse energy of 120 μ\muJ and a repetition rate of 525 kHz. High harmonics are generated by focusing the fundamental beam into a synthetic fused-silica hollow-core fiber waveguide (XUUS-5, KM Labs) filled with xenon gas. The resulting extreme-ultraviolet (XUV) radiation is monochromated to a photon energy of 22.8 eV, corresponding to the highest harmonic generated with appreciable flux under these conditions. The monochromated XUV beam is focused onto the sample with a spot size of 200 μ\mum and serves as the photoemission probe for both conventional ARPES and two-electron coincidence measurements. Photoelectrons are detected using two angle-resolved time-of-flight analyzers (ARTOF-2, Scienta Omicron) positioned perpendicular to each other. The sample temperature was controlled between 8 K and 300 K using a closed-cycle helium cryostat (ARS-4HW compressor and DE-202SB cold head, Advanced Research Systems). A six-axis manipulator (CryoArc) coupled to the cold head provided precise control of the sample position and orientation, with translational and angular resolutions of 1 µm and 10−310^{-3} rad, respectively.

Correlation measure: Each detected photoelectron is characterized by its energy EiE_i and in-plane momentum ki\mathbf{k}_i, collectively denoted by ξi=(Ei,ki)\boldsymbol{\xi}_i = (E_i,\mathbf{k}_i), where i=1,2i = 1,2 labels the detector. In our 2e-ARPES measurements, EiE_i is defined relative to the Fermi level, as determined from conventional ARPES measurements. The same-pulse coincidence distribution, I(2)(ξ1,ξ2)I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2), counts events in which one electron is detected in each analyzer following the same laser pulse. These events include true double photoemission, in which a single photon induces the emission of two electrons, and accidental coincidences arising from independent single-electron photoemission events within the same laser pulse.

For statistically independent single-electron emission, the expected accidental-coincidence distribution factorizes as

Iacc(2)(ξ1,ξ2)=I(1)(ξ1)I(1)(ξ2),(1)I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = I^{(1)}(\boldsymbol{\xi}_1)I^{(1)}(\boldsymbol{\xi}_2), \tag*{(1)}

where I(1)(ξi)I^{(1)}(\boldsymbol{\xi}_i) denotes the single-electron intensity, normalized such that its product gives the expected number of accidental coincidences. Experimentally, Iacc(2)I_{\mathrm{acc}}^{(2)} is estimated by pairing electrons detected in different laser pulses, as described below. Because electrons from different pulses cannot originate from the same microscopic emission event, this provides a direct estimate of the uncorrelated background.

We define the two-electron correlation measure

P(2)(ξ1,ξ2)=[I(2)(ξ1,ξ2)−Iacc(2)(ξ1,ξ2)]2I(2)(ξ1,ξ2)+Iacc(2)(ξ1,ξ2)+ϵ,(2)P^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \frac{\left[I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) - I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2)\right]^2}{I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) + I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) + \epsilon}, \tag*{(2)}

where ϵ>0\epsilon> 0 is a small regularization parameter that stabilizes bins containing very few counts. The numerator represents the squared coincidence excess above the accidental background, while the denominator provides a normalization based on the total counting intensity. Thus, P(2)P^{(2)} characterizes the strength of correlations beyond those expected from independent photoemission. The features analyzed in this work correspond to a positive excess of same-pulse coincidences before the difference is squared. Further details of the pulse-offset subtraction and the connection between the coincidence rate and the two-electron photoemission probability are provided in the Supplementary Information.

Coincidence data analysis: Photoelectrons were recorded using two angle-resolved time-of-flight analyzers, synchronized to the laser repetition clock. Each detection event was assigned a laser-pulse index nn, allowing electron pairs to be identified according to their pulse of origin.

For each energy- and momentum-resolved bin (ξ1,ξ2)(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2), we construct the pulse-offset coincidence histogram:

Ml(ξ1,ξ2)=∑nc1,n(ξ1)c2,n−l(ξ2),(3)M_l(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \sum_n c_{1,n}(\boldsymbol{\xi}_1)c_{2,n-l}(\boldsymbol{\xi}_2), \tag*{(3)}

where ci,n(ξi)c_{i,n}(\boldsymbol{\xi}_i) equals one when detector ii records exactly one electron in bin ξi\boldsymbol{\xi}_i from pulse nn, and zero otherwise. The integer ll denotes the pulse offset between the two detectors, with the sum restricted to valid pulse indices.

The measured same-pulse coincidence distribution is

I(2)(ξ1,ξ2)=M0(ξ1,ξ2).(4)I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = M_0(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2). \tag*{(4)}

The accidental-coincidence distribution is estimated by averaging the histograms obtained from the two nearest-neighbor pulse offsets:

Iacc(2)(ξ1,ξ2)=M+1(ξ1,ξ2)+M−1(ξ1,ξ2)2.(5)I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \frac{M_{+1}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) + M_{-1}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2)}{2}. \tag*{(5)}

Because the detected electron time-of-flight window is shorter than the laser repetition period, electrons assigned to different pulses cannot originate from the same microscopic photoemission event. Provided that the pulse-to-pulse detection statistics are stationary and independent, the offset histograms therefore estimate the accidental background. The correlation measure P(2)(ξ1,ξ2)P^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) is then calculated using Eq. 2. Additional experimental checks and details of the background estimation are provided in the Supplementary Information.

Bootstrapping: Statistical uncertainties in P(2)P^{(2)} are estimated using bootstrap resampling [38]. The same-pulse and pulse-offset coincidence samples are resampled independently with replacement. For each bootstrap realization, the distributions I(2)(ξ1,ξ2)I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) and Iacc(2)(ξ1,ξ2)I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) are reconstructed, and the corresponding P(2)(ξ1,ξ2)P^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) is calculated using Eq. 2. This procedure is repeated to generate a bootstrap distribution of P(2)P^{(2)} in each (ξ1,ξ2)(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) bin. The standard deviation of this distribution is taken as the statistical uncertainty. Further details are provided in the Supplementary Information.

Data availability: The data in this manuscript are available from the corresponding author upon reasonable request.

Corresponding author: Correspondence to Fahad Mahmood ([email protected]).

ILLUSTRATIVE SIMULATION OF THE COOPER-PAIR EMISSION PEAK

To illustrate the characteristic two-electron emission peak expected from the superconducting condensate, we calculate the pair-photoemission signal for a dd-wave superconductor with zero pair momentum (k1=−k2\mathbf{k}_1=-\mathbf{k}_2). This calculation provides the illustrative joint energy distribution shown in Figure 1b of the Main Text. The Bi-2212 Fermi surface is modeled using the following tight-binding dispersion

ξk=−2t1(cos⁡kx+cos⁡ky)−4t2cos⁡kxcos⁡ky−2t3(cos⁡2kx+cos⁡2ky)−4t4(cos⁡2kxcos⁡ky+cos⁡2kycos⁡kx)−μ(6)\xi_{\mathbf{k}}=-2t_1(\cos k_x+\cos k_y)-4t_2\cos k_x\cos k_y-2t_3(\cos2k_x+\cos2k_y)-4t_4(\cos2k_x\cos k_y+\cos2k_y\cos k_x)-\mu \tag*{(6)}

where t1=368t_1=368 meV, t2=−0.342t1t_2=-0.342t_1, t3=0.11t1t_3=0.11t_1, t4=−0.087t1t_4=-0.087t_1 are the hopping amplitudes and μ=440\mu=440 meV is the chemical potential. The superconducting order parameter is given by

Δk=Δ02(cos⁡kx−cos⁡ky)(7)\Delta_{\mathbf{k}}=\frac{\Delta_0}{2}(\cos k_x-\cos k_y) \tag*{(7)}

where Δ0=30\Delta_0=30 meV. Suppose that the two photoelectrons carry momenta k1=(kx,ky)\mathbf{k}_1=(k_x,k_y) and k2=(−kx,−ky)\mathbf{k}_2=(-k_x,-k_y) and opposite spin, the pair photoemission rate is given by [1]:

PSC(2)(kx,ky;Γ)=2π[1πΓ(ℏω−2ϵk)2+Γ2]∣∫d2p(2π)2V(p−k)ℏω−ϵp−Ep+iδΔp2Ep∣2(8)P_{\mathrm{SC}}^{(2)}(k_x,k_y;\Gamma)=2\pi\left[\frac{1}{\pi}\frac{\Gamma}{(\hbar\omega-2\epsilon_k)^2+\Gamma^2}\right]\left|\int\frac{\mathrm{d}^2p}{(2\pi)^2}\frac{V(\mathbf{p}-\mathbf{k})}{\hbar\omega-\epsilon_p-E_p+\mathrm{i}\delta}\frac{\Delta_p}{2E_p}\right|^2 \tag*{(8)}

where ℏω\hbar\omega is the photon energy, ϵk=ℏ2k22mea02+W\epsilon_k=\frac{\hbar^2 k^2}{2m_ea_0^2}+W is the sum of the photoelectron’s kinetic energy and work function, and Ep=ξp2+Δp2E_p=\sqrt{\xi_p^2+\Delta_p^2} is the dispersion of the superconducting state. To account for finite momentum broadening in the simulated pair photoemission distribution, we introduce Lorentzian broadening of the theoretical result. We define

L(x;γ)=1πγx2+γ2(9)L(x;\gamma)=\frac{1}{\pi}\frac{\gamma}{x^2+\gamma^2} \tag*{(9)}

where γ\gamma is the Lorentzian half-width at half maximum. The broadened pair photoemission rate is then

Pconv(2)(k1,k2)=L(k1x+k2x;γsum)L(k1y+k2y;γsum)∑nL(k1x−k2x−2kx(n);γdiff)L(k1y−k2y−2ky(n);γdiff)PSC(2)(kx(n),ky(n);Γ)(10)P_{\mathrm{conv}}^{(2)}(\mathbf{k}_1,\mathbf{k}_2)=L(k_{1x}+k_{2x};\gamma_{\mathrm{sum}})L(k_{1y}+k_{2y};\gamma_{\mathrm{sum}}) \sum_n L(k_{1x}-k_{2x}-2k_x^{(n)};\gamma_{\mathrm{diff}})L(k_{1y}-k_{2y}-2k_y^{(n)};\gamma_{\mathrm{diff}})P_{\mathrm{SC}}^{(2)}(k_x^{(n)},k_y^{(n)};\Gamma) \tag*{(10)}

Here, the summation is performed over a discrete momentum grid within a box of width 0.2a0−10.2a_0^{-1} centered at the antinodal point of the dd-wave Fermi surface to reduce computational cost.

We introduce distinct Lorentzian broadening widths, γsum\gamma_{\mathrm{sum}} and γdiff\gamma_{\mathrm{diff}}, for the sum momentum k1+k2\mathbf{k}_1+\mathbf{k}_2 and difference momentum k1−k2\mathbf{k}_1-\mathbf{k}_2, respectively. Allowing these widths to vary independently provides a phenomenological description of anisotropic momentum broadening in the simulated two-electron distribution.

Experimental details and single-electron ARPES characterization

Sample information

Optimally doped Bi2Sr2CaCu2O8+δ\mathrm{Bi}_2\mathrm{Sr}_2\mathrm{CaCu}_2\mathrm{O}_{8+\delta} (Bi-2212) single crystals, with cross-sectional areas of approximately 11–4 mm24\ \mathrm{mm}^2, were mounted on copper sample plates using conductive silver epoxy to provide thermal and electrical contact with the manipulator. Samples were cleaved in situ under ultrahigh vacuum by removing a post attached to the crystal, exposing a fresh surface parallel to the crystallographic abab plane. Before acquiring ARPES or two-electron ARPES (2e-ARPES) data, we mapped the photoemission intensity over the exposed surface to select a spatially uniform measurement region. The incident beam had a spot diameter of approximately 200 μm200\ \mu\mathrm{m}, substantially smaller than the lateral dimensions of the crystals.

Tight-binding dispersion and accessible momentum regions

The measured Fermi surface (Figure 2 of the Main Text) is consistent with the published tight-binding parametrization of Bi-2212 [2]. For comparison with the experimental momentum coverage, we show the bonding-band dispersion and omit superstructure replicas [3, 4]. The highlighted regions in Figure 2 (Main Text) indicate the in-plane momenta accessible to the two detectors for the geometries summarized in Supplementary Table 1.

The magnitude of the photoelectron momentum parallel to the sample surface is determined from its kinetic energy and emission angle:

k∥=2meEkℏ2sin⁡θ,(11)k_{\parallel} = \sqrt{\frac{2m_e E_k}{\hbar^2}}\sin\theta, \tag*{(11)}

where mem_e is the electron mass, EkE_k is the photoelectron kinetic energy, ℏ\hbar is the reduced Planck constant and θ\theta is the emission angle relative to the surface normal. A 15∘15^\circ angular field of view was assumed for each detector. For conventional ARPES, the tabulated kinetic energy corresponds to emission from the Fermi level, Ek=hν−ΦE_k = h\nu- \Phi, where hνh\nu is the photon energy and Φ\Phi is the sample work function. For 2e-ARPES, Ek=hν/2−ΦE_k = h\nu/2 - \Phi denotes the nominal kinetic energy of each electron for equal energy sharing and zero pair binding energy.

RegionDetectorEkE_kMeasurementSurface normal
12hν−Φh\nu- \PhiARPESx^\hat{x}
21hν−Φh\nu- \PhiARPES12(x^−y^)\frac{1}{\sqrt{2}}(\hat{x}-\hat{y})
31hν−Φh\nu- \PhiARPESx^\hat{x}
42hν2−Φ\frac{h\nu}{2}-\Phi2e-ARPESx^\hat{x}
51hν2−Φ\frac{h\nu}{2}-\Phi2e-ARPESx^\hat{x}

Table 1. Supplementary Table 1. Experimental geometries used to determine the momentum regions in Fig. 2 of the main text. The detector axes lie along (x^−y^)/2(\hat{x}-\hat{y})/\sqrt{2} (detector 1) and (x^+y^)/2(\hat{x}+\hat{y})/\sqrt{2} (detector 2); the incident beam propagates along −x^-\hat{x}. All directions are specified in the laboratory coordinate system. The 2e-ARPES kinetic energies correspond to equal energy sharing at zero pair binding energy.

Gap structure of Bi-2212

The superconducting gap of our optimally-doped Bi-2212 samples was characterized by conventional ARPES. Supplementary Fig. S1a shows energy-distribution curves (EDCs) at points along the Fermi surface, parametrized by the Fermi-surface angle θ\theta defined in the figure. The leading-edge position shifts to higher binding energy away from the nodal direction (θ=π/4\theta=\pi/4), consistent with the opening of an anisotropic superconducting gap.

To quantify this shift, each EDC was fitted phenomenologically with a Fermi-Dirac edge multiplied by a linear spectral background and convolved with a Gaussian broadening function:

I(E)=[A+BE1+exp⁡(E−μkBT)]∗12πσexp⁡(−E22σ2),(12)I(E) = \left[\frac{A+BE}{1+\exp\left(\frac{E-\mu}{k_{\mathrm{B}}T}\right)}\right] * \frac{1}{\sqrt{2\pi}\sigma}\exp\left(-\frac{E^2}{2\sigma^2}\right), \tag*{(12)}

where EE is electron energy relative to the experimental energy reference, TT is the sample temperature, AA and BB describe the locally linear prefactor, μ\mu is the fitted edge position and σ\sigma is the Gaussian broadening width. The symbol ∗\ast denotes convolution over energy. The parameter μ\mu is used here only to track the leading-edge position; it is not interpreted as a momentum-dependent thermodynamic chemical potential. Fits based on a Dynes spectral function [5] were not sufficiently stable at the available energy resolution and counting statistics.

Taking the difference between μ\mu and the Fermi level at the node, EFE_F, we obtain an estimate for the gap Δ\Delta. Its angular dependence is consistent with the expected dd-wave gap anisotropy of optimally doped Bi-2212 (Supplementary Fig. S1b) [2, 6, 7].

Single-electron ARPES energy distribution curves

(a)

Angular dependence of the gap estimate

(b)

Figure S1. Single-electron ARPES characterization of optimally doped Bi-2212 at 33 K. a EDCs recorded at positions along the Fermi surface, indicated by the inset. Each EDC is normalized to its mean intensity over the energy interval [−200,200-200, 200] meV. b, Angular dependence of the gap estimate Δ\Delta, determined from fits to the EDCs as described in the text. The angle θ\theta is defined in a. Error bars represent 68% confidence intervals of the fitted estimates.

2E-ARPES

Coincidence Algorithm and Data Analysis

Photoelectrons were recorded using two angle-resolved time-of-flight analyzers, synchronized to the laser repetition clock. Each detection event was assigned a laser-pulse index nn, an energy EiE_i, and an in-plane momentum ki\mathbf{k}_i, where i=1,2i = 1, 2 labels the detector. We collectively denote the energy and momentum coordinates by ξi=(Ei,ki)\boldsymbol{\xi}_i = (E_i,\mathbf{k}_i).

A same-pulse coincidence was registered when each detector recorded exactly one electron originating from the same laser pulse. Such events can arise from two distinct processes: true double photoemission, in which a single photon induces the emission of two electrons, or accidental coincidences, in which independent single-electron photoemission events occur within the same laser pulse. Since both processes contribute to the measured coincidence distribution, the accidental background must be determined independently to isolate the correlated two-electron signal.

The measured same-pulse coincidence distribution is denoted by I(2)(ξ1,ξ2)I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2). For statistically independent single-electron emission, the expected accidental-coincidence distribution factorizes as

Iacc(2)(ξ1,ξ2)=I(1)(ξ1)I(1)(ξ2),(13)I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = I^{(1)}(\boldsymbol{\xi}_1)I^{(1)}(\boldsymbol{\xi}_2), \tag*{(13)}

where I(1)(ξi)I^{(1)}(\boldsymbol{\xi}_i) is the appropriately normalized single-electron intensity in detector ii.

Experimentally, the accidental background is estimated directly from detections associated with different laser pulses. We define the pulse-offset coincidence histogram

Ml(ξ1,ξ2)=∑nc1,n(ξ1)c2,n−l(ξ2),(14)M_l(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \sum_n c_{1,n}(\boldsymbol{\xi}_1)c_{2,n-l}(\boldsymbol{\xi}_2), \tag*{(14)}

where ci,n(ξi)c_{i,n}(\boldsymbol{\xi}_i) equals one when detector ii records exactly one electron in bin ξi\boldsymbol{\xi}_i from pulse nn, and zero otherwise. The integer ll specifies the pulse offset, and the sum includes only valid pulse indices.

For l=0l = 0, events are paired using the same laser pulse in both detectors, giving

I(2)(ξ1,ξ2)=M0(ξ1,ξ2).(15)I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = M_0(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2). \tag*{(15)}

For nonzero ll, an event from pulse nn in detector 1 is paired with an event from pulse n−ln-l in detector 2. Because the two electrons originate from different laser pulses, they cannot result from the same microscopic double-photoemission event. These offset distributions therefore provide an estimate of the accidental background, provided the detection statistics are stationary and independent between pulses.

For the analysis presented in the main text, we average the two nearest-neighbor pulse offsets:

Iacc(2)(ξ1,ξ2)=12[M+1(ξ1,ξ2)+M−1(ξ1,ξ2)].(16)I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \frac{1}{2}\left[M_{+1}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) + M_{-1}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2)\right]. \tag*{(16)}

The difference I(2)(ξ1,ξ2)−Iacc(2)(ξ1,ξ2)I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) - I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) then gives the measured coincidence excess above the accidental background. Supplementary Fig. S2 illustrates the pulse-matching procedure. The synchronous histogram M0M_0 is constructed by identifying pulses for which both detectors recorded an electron, whereas M+1M_{+1} is obtained by pairing detections from pulse nn in detector 1 with those from pulse n−1n-1 in detector 2.

As in the Methods section of the main text, we characterize the correlated two-electron emission using

P(2)(ξ1,ξ2)=[I(2)(ξ1,ξ2)−Iacc(2)(ξ1,ξ2)]2I(2)(ξ1,ξ2)+Iacc(2)(ξ1,ξ2)+ϵ,(17)P^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \frac{\left[I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) - I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2)\right]^2}{I^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) + I_{\mathrm{acc}}^{(2)}(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) + \epsilon}, \tag*{(17)}

where ϵ>0\epsilon> 0 is a small regularization parameter that prevents numerical instabilities in bins containing very few counts. The numerator measures the squared excess of same-pulse coincidences over the accidental background, while the denominator weights this excess by the total counting intensity.

To assess the accidental-coincidence background independently, we also examine the combined histogram obtained from six nonzero pulse offsets,

IB(ξ1,ξ2)=∑l=−3l≠03Ml(ξ1,ξ2).(18)I_B(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2) = \sum_{\substack{l=-3 \\ l\ne0}}^{3} M_l(\boldsymbol{\xi}_1,\boldsymbol{\xi}_2). \tag*{(18)}

Supplementary Fig. S3 shows this distribution for the 8.5 K dataset used in the main text. The background exhibits no discernible structure beyond statistical counting fluctuations.

Figure S2. Pulse-matching algorithm for 2e-ARPES. For this generated data, detectors 1 and 2 (D1D_1 and D2D_2) are filled in for pulse nn if there was a detected event. M0M_0 is found by identifying the pulses labeled nn where both detectors observed an event. M+1M_{+1} is found by identifying cases where there was an event from pulse nn in detector 1 and event from pulse n−1n-1 in detector 2.

Figure S3. 2e-ARPES background. Intensity plot of IB(ξ1,ξ2)=∑l=−3l≠03Ml(ξ1,ξ2)I_B(\xi_1,\xi_2)=\sum_{\substack{l=-3\\l\ne0}}^{3}M_l(\xi_1,\xi_2) for the 8.5 K dataset in the main text. Intensity modulations can be attributed to shot noise.

Optimization of P(2)P^{(2)}

As established in previous sections, the sensitivity of 2e-ARPES measurements is determined by the number of correlated two-electron events relative to the statistical fluctuations of the accidental-coincidence background. Consequently, the value of P(2)P^{(2)} depends both on sample-dependent photoemission probabilities and on experimental parameters. The role of the latter must be characterized in order to maximize P(2)P^{(2)}.

We model the detection statistics using three probabilities: PCP_C, the probability of detecting a correlated two-electron photoemission event per incident photon; PSP_S, the probability of detecting a single-electron photoemission event in either analyzer per incident photon; and PBP_B, the probability of a detector dark count within one laser-pulse window. We assume comparable detection probabilities for the two analyzers and PC≪PSP_C \ll P_S. Additionally, PCP_C and PSP_S are inherently functions of ξ1\xi_1 and ξ2\xi_2.

For a Poisson distribution of incident photons with mean λ\lambda per laser pulse, the probability of detecting a true two-electron event is, in the low-count-rate limit,

Ptrue0≃λPC.(19)P_{\mathrm{true}}^{0} \simeq\lambda P_{\mathrm{C}}. \tag*{(19)}

Accidental coincidences arise when the two analyzers independently detect single photoelectrons or dark counts within the same laser pulse. Their probability is approximately

Pacc≃(λPS+PB)2.(20)P_{\mathrm{acc}} \simeq\left(\lambda P_{\mathrm{S}} + P_{\mathrm{B}}\right)^2. \tag*{(20)}

Here, PBP_{\mathrm{B}} scales inversely with the laser repetition rate ff for a fixed detector dark-count rate. These expressions apply in the regime of low detector occupancy, where multiple detections within a single laser pulse are negligible.

The finite lifetime of the freshly cleaved sample surface introduces an additional constraint on the acquisition time. We model the decrease in the correlated photoemission probability as

Ptrue(t)=Ptrue0e−t/τ,(21)P_{\mathrm{true}}(t) = P_{\mathrm{true}}^{0} e^{-t/\tau}, \tag*{(21)}

where tt is the elapsed acquisition time and τ\tau is the characteristic decay time, which depends on the sample and vacuum conditions. The single-electron background is taken to remain approximately constant over the acquisition.

For an acquisition of duration tt, the expected number of correlated events is therefore

Ntrue(t)=Ptrue0fτ(1−e−t/τ),(22)N_{\mathrm{true}}(t) = P_{\mathrm{true}}^{0} f \tau\left(1-e^{-t/\tau}\right), \tag*{(22)}

while the expected number of accidental coincidences is

Nacc(t)=Paccft.(23)N_{\mathrm{acc}}(t) = P_{\mathrm{acc}} f t. \tag*{(23)}

Assuming Poisson counting statistics, the variance of the background-subtracted coincidence signal is

σ2=Ntrue+Nacc(1+1γ),(24)\sigma^2 = N_{\mathrm{true}} + N_{\mathrm{acc}}\left(1+\frac{1}{\gamma}\right), \tag*{(24)}

where γ\gamma is the number of independent nonzero pulse offsets used to estimate Iacc(2)I_{\mathrm{acc}}^{(2)}. The factor 1+1/γ1+1/\gamma accounts for fluctuations in both the same-pulse coincidence distribution I(2)I^{(2)} and the estimated accidental background. For the nearest-neighbor subtraction used in this work, γ=2\gamma=2.

Combining these expressions gives

P(2)(t)=[Ptrue0fτ(1−e−t/τ)]2Ptrue0fτ(1−e−t/τ)+Pacc(1+1γ)ft.(25)P^{(2)}(t) = \frac{\left[P_{\mathrm{true}}^{0}f\tau\left(1-e^{-t/\tau}\right)\right]^2}{P_{\mathrm{true}}^{0}f\tau\left(1-e^{-t/\tau}\right)+P_{\mathrm{acc}}\left(1+\frac{1}{\gamma}\right)ft}. \tag*{(25)}

In the limit of low photon occupancy and negligible sample degradation, this expression reduces to the squared signal-to-noise ratio discussed in Ref. [8].

Supplementary Figure S4 shows the calculated P(2)P^{(2)} as a function of the laser repetition rate ff and the mean photon number per pulse λ\lambda, using representative experimental parameters. For each combination of ff and λ\lambda, the acquisition time is chosen to maximize P(2)P^{(2)}. The calculations illustrate the advantage of increasing the repetition rate while maintaining a fixed average photon flux. Under these conditions, the correlated event rate remains approximately constant, whereas the probability of accidental coincidences decreases because fewer photoelectrons are generated per pulse.

Figure S4. Simulated P(2)P^{(2)}. a P(2)P^{(2)} integrated over ξ1\xi_1 and ξ2\xi_2 as a function of the average number of photons per pulse λ\lambda and the laser repetition rate with PC=3×10−8P_C = 3 \times10^{-8}, PS=1×10−3P_S = 1 \times10^{-3}, and τ=3×104\tau= 3 \times10^{4} s. For each pixel, the acquisition time tt is chosen such that it maximizes P(2)P^{(2)}, which is our procedure after a measurement. Due to experimental constraints, the measurement must be performed to the left of the black-dashed line. The value of P(2)P^{(2)} along this line is given in b.

In practice, the accessible photon flux is limited by the detector count rate, while the available extreme-ultraviolet photon flux depends on the efficiency of high-harmonic generation. The repetition rate of f=525 kHzf=525\ \mathrm{kHz} used in the main-text measurements represents a compromise between these constraints, as the high-harmonic photon flux decreases sharply at higher repetition rates in the present setup.

Pulse independence

The pulse-offset coincidence analysis requires that detections associated with successive laser pulses are statistically independent. One potential source of ambiguity arises from the spread in electron time-of-flights. Slow photoelectrons emitted by one laser pulse could overlap with faster photoelectrons emitted by a subsequent pulse. Supplementary Figure S5 shows that the detected electron time-of-flight window is shorter than the laser repetition period, tν=1/ft_{\nu}=1/f, where f=525f = 525 kHz is the repetition rate used in the experiment. The electronic dead time following each detection is approximately 150 ns, also substantially shorter than the laser repetition period.

Figure S5. Pulse independence. a Histogram of detected electron time-of-flights. The x-axis span one laser period for the repetition rate f=525f = 525 kHz used in the experiment. b Coincidence counts MlM_l as a function of pulse offset ll (as defined in the Methods section in the main text). The grey region encapsulates 2σ2\sigma, where σ=N\sigma= \sqrt{N}. This is representative of Poisson counting statistics, and NN is the average of all data points. The reduced χ2\chi^2, calculated as χred2=1N(p−1)∑l=1p(Ml−N)2\chi_{\mathrm{red}}^2 = \frac{1}{N(p-1)}\sum_{l=1}^{p}(M_l-N)^2 where pp is the number of offsets, is approximately equal to 1.

A second consideration is the possibility of systematic pulse-to-pulse intensity variations. The laser system employs two Pockels cells for pulse amplification and pulse selection, respectively. The pulse picker reduces the repetition rate by transmitting every mmth pulse, where m≥1m \ge1 is an integer, while maintaining approximately constant pulse energy. However, pulse selection can introduce periodic intensity modulations, potentially producing correlations between detections associated with different laser pulses.

To test for such effects, we examine the coincidence counts MlM_l as a function of the nonzero pulse offset ll, as defined in the preceding section. Here, MlM_l denotes the total coincidence counts integrated over the measured energy and momentum regions. As shown in Supplementary Fig. S5b, the counts for l>0l > 0 are consistent with a constant mean and Poisson counting fluctuations, with a reduced chi-squared value of χred2=0.891\chi_{\mathrm{red}}^2 = 0.891. This agreement supports the assumption of statistically independent pulse-to-pulse detections underlying the accidental-coincidence background subtraction.

Figure generation

Figures were generated using Python. Data processing, visualization, and figure formatting were performed using the python package Matplotlib and other standard Python packages and libraries, as appropriate. Constant intensity contour plots of P(2)P^{(2)} were made using Matplotlib’s built in Contour method, where lines are drawn connecting points of equal intensity. P(2)P^{(2)} was calculated using the same method as outlined in the Coincidence Algorithm and Data Analysis section of this supplement and normalized by the total intensity. Contour lines of matching color in each plot represent the same relative intensity of P(2)P^{(2)}. The data used to generate the figures are available in the Supplementary Information Source Data files.

Bootstrapping

Statistical uncertainties in the two-electron correlation measure P(2)P^{(2)} were determined using bootstrap resampling [9]. The same-pulse and pulse-offset coincidence samples were independently resampled with replacement. For each bootstrap realization, the distributions I(2)(ξ1,ξ2)I^{(2)}(\xi_1,\xi_2) and Iacc(2)(ξ1,ξ2)I_{\mathrm{acc}}^{(2)}(\xi_1,\xi_2) were reconstructed, and the corresponding P(2)(ξ1,ξ2)P^{(2)}(\xi_1,\xi_2) was calculated using Eq. (17). Repeating this procedure generates a distribution of P(2)P^{(2)} values for each energy- and momentum-resolved bin.

It is standard to represent the error of a bootstrapped quantity xx as the product of the standard error and a critical TT value [9]. For example, a critical TT value of 1 is associated with a 68 percent confidence interval for normally distributed data points. It follows that the lower error bar for each data point should go to the 16th percentile and the upper error bar should go to the 84th percentile of the distribution of bootstrapped means. To find these values for a discrete set of data, we can use the expression from [10]

Q^=(1−g)xj+gxj+1,(26)\hat{Q} = (1-g)x_j + gx_{j+1}, \tag*{(26)}

where j=[qn]j = \left[qn\right] and g=qn−jg = qn-j. Here, the [x]\left[x\right] operator returns the largest integer less than xx. Also, NN is the number of points to average over, n=N−1n = N - 1, and qq is the percentile divided by 100. For example, to obtain the 16th percentile, q=0.16q = 0.16.

Broadening of the E1=E2=0E_1 = E_2 = 0 peak due to intermediate state scattering

To understand the broadening in the (E1,E2)(E_1,E_2) plane of the experimentally observed peak in the photo-electron counting rate P(2)P^{(2)} which is centered at E1=E2=0E_1 = E_2 = 0, we recall that this peak arises from the emission of two photo-electrons from the same Cooper pair. This contribution to P(2)P^{(2)} is referred to as PSC(2)P^{(2)}_{SC} to distinguish it from the contribution, P2cp(2)P_{2cp}^{(2)}, that is located at higher energies and arises from photo-electrons emitted by two different Cooper pairs. PSC(2)P^{(2)}_{SC} arises from two consecutive processes: the emission of a first photo-electron upon absorption of a photon, and the subsequent scattering (EELS-like) process between the emitted photo-electron and a conduction electron, which leads to the emission of a second photo-electron, as shown in Figure S6. For PSC(2)P^{(2)}_{SC}, the emitted photo-electrons originate from the same Cooper pair, and thus conduction electrons with momenta k\mathbf{k} and −k-\mathbf{k}, and thus zero center-of-mass momentum. We assume that the scattering process between the photo-electron and the conduction electron is mediated by a (screened) Coulomb interaction. This entire process is then described by the Hamiltonian

Two-step and intermediate state scattering processes

Figure S6. (a) Two-step scattering process describing the absorption of a photon and the emission of two photo-electrons. (b) Same process as in (a), but including an intermediate state scattering process in which the photo-electron scatters a conduction electron back into the conduction band. (c) Same as in (a), but with multiple intermediate state scattering processes. (d) Accounting for the strong Coulomb interaction in the conduction band, the intermediate state scattering process can be considered to be mediated by spin or charge collective modes.

Hsc=∑k,σ,νγν(q)dk+q,σ†ck,σ(aq,ν+a−q,ν†)+∑k,p,q,α,βV(q)dk+q,α†dp−q,β†dp,βck,α+h.c.(27)H_{sc} = \sum_{\mathbf{k},\sigma,\nu} \gamma_{\nu}(\mathbf{q})d^{\dagger}_{\mathbf{k}+\mathbf{q},\sigma}c_{\mathbf{k},\sigma}\left(a_{\mathbf{q},\nu}+a^{\dagger}_{-\mathbf{q},\nu}\right) + \sum_{\mathbf{k},\mathbf{p},\mathbf{q},\alpha,\beta} V(\mathbf{q})d^{\dagger}_{\mathbf{k}+\mathbf{q},\alpha}d^{\mathbf{\dagger}}_{\mathbf{p}-\mathbf{q},\beta}d_{\mathbf{p},\beta}c_{\mathbf{k},\alpha} + h.c. \tag*{(27)}

Here, γν(q)\gamma_{\nu}(\mathbf{q}) is the effective electron-photon dipole interaction, dk,σ†d^{\dagger}_{\mathbf{k},\sigma} (ck,σc_{\mathbf{k},\sigma}) creates (destroys) a photo-electron (conduction electron) with momentum k\mathbf{k} and spin σ\sigma, and V(q)∼[q2+κ2]−1V(\mathbf{q}) \sim[q^2+\kappa^2]^{-1} is the Fourier transform of the (screened) Coulomb interaction, with κ−1\kappa^{-1} being the screening length.

However, the presence of a Coulomb interaction between the first photo-electron and the conduction electron can not only lead to the emission of a photo-electron, but also to a scattering process in which the conduction electron is scattered back into its band – a process we refer to as intermediate state scattering –, which is described by the Hamiltonian

Hc=∑k,p,q,α,βV(q)dk+q,α†cp−q,β†dk,βcp,α+h.c.(28)H_c = \sum_{\mathbf{k},\mathbf{p},\mathbf{q},\alpha,\beta} V(\mathbf{q})d^{\dagger}_{\mathbf{k}+\mathbf{q},\alpha}c^{\dagger}_{\mathbf{p}-\mathbf{q},\beta}d_{\mathbf{k},\beta}c_{\mathbf{p},\alpha} + h.c. \tag*{(28)}

Such a process, as shown in Figure S6 transfers momentum, and hence energy, between the photo-electron and the conduction band. As a result, after the emission of the second photo-electron, the two photo-electrons do not possess a zero center-of-mass momentum, and thus now exhibit a finite distribution in the (E1,E2)(E_1,E_2) plane. In general, this intermediate state scattering process can occur multiple time before the final emission of second photo-electron, as

shown in Figure S6. Moreover, since the emergence of collective spin or charge modes in the cuprate superconductors arises from Coulomb scattering within the conduction band, the inclusion of higher order scattering events in the conduction band allows one to describe the scattering between the photo-electron and conduction band as mediated by a collective spin or charge mode, as shown in Figure S6 (this can be seen, for example, by representing the collective mode as an infinite series of bubble diagrams in the random-phase approximation). This implies that the distribution of the photo-electron energies in the (E1,E2)(E_1,E_2) plane is a direct probe for collective modes in the cuprate superconductors. The detailed investigation of how these collective modes affect the intensity distribution of PSC(2)P_{SC}^{(2)} in the (E1,E2)(E_1,E_2) plane is reserved for future work.

Broadening of E−E^{-}: Relative-Time Dynamics and Dephasing

To construct our model, we use the general approach of [11]. In particular, we start by considering the low-energy electronic states, described by the operator ckσ,αc_{\mathbf{k}\sigma,\alpha}, the intermediate high-energy electronic states, described by dqσd_{\mathbf{q}\sigma}, and the photons, described by ap′γa_{\mathbf{p}'\gamma}. Here, the subscript α\alpha denotes the set of internal degrees of freedom of the low-energy electrons (like band or orbital), σ\sigma is the spin, and γ\gamma is the photon polarization. While the low-energy states are those that form the superconducting condensate below TcT_c, the high-energy electrons are connected to, but are not identified with, the asymptotic photoelectrons with momentum p\mathbf{p} and spin σ\sigma measured in the detector, as explained in Ref. [11]. The total Hamiltonian can then be written as H=H0+H1H = H_0 + H_1, with H0H_0 denoting the terms of the Hamiltonian corresponding to the three independent sectors,

H0=Hph+Hlow+Hhigh,(29)H_0 = H_{\mathrm{ph}} + H_{\mathrm{low}} + H_{\mathrm{high}}, \tag*{(29)}

and H1H_1 denoting the terms of the Hamiltonian corresponding to the interactions between the three independent sectors,

H1=Hac+Hcd+Hacd.(30)H_1 = H_{ac} + H_{cd} + H_{acd}. \tag*{(30)}

We emphasize that H0H_0 does not correspond to a free Hamiltonian, as interactions within each sector are allowed. To proceed, we approximate the photon and high-energy Hamiltonian terms HphH_{\mathrm{ph}} and HhighH_{\mathrm{high}} as free (i.e., they contain only one-body operators) but consider the full interacting cc-electron Hamiltonian HcH_c, and define Hlow=HcH_{\mathrm{low}} = H_c. Importantly, the intermediate dd-electron modes are distinguished from the asymptotic outgoing photoelectron states. Therefore, even without specifying a microscopic mechanism, we can encode the process by which a dd-electron with momentum p′\mathbf{p}' and spin σ′\sigma' is converted into a photoelectron with momentum p\mathbf{p} and spin σ\sigma through a matrix element Mσσ′(p,p′)M_{\sigma\sigma'}(\mathbf{p},\mathbf{p}').

Our goal is to compute the transition operator for the single-photon-absorption channel, starting from a known coherent photon probe state and the high-energy-electron vacuum at the initial time tit_i. We project at the final time tft_f onto two asymptotic outgoing photoelectrons with momenta p1\mathbf{p}_1, p2\mathbf{p}_2 and spins σ1\sigma_1, σ2\sigma_2. Formally, this transition operator is given by:

Aσ1σ2μ(p1,p2;tf,ti)≡out⟨p1σ1,p2σ2∣γ⟨μ∣U(tf,ti)∣μ0⟩γ0∣0⟩d.(31)\mathcal{A}^{\mu}_{\sigma_1\sigma_2}(\mathbf{p}_1,\mathbf{p}_2;t_f,t_i) \equiv{}_{\mathrm{out}}\langle\mathbf{p}_1\sigma_1,\mathbf{p}_2\sigma_2|{}_{\gamma}\langle\mu|U(t_f,t_i)|\mu_0\rangle_{\gamma_0}|0\rangle_d. \tag*{(31)}

Here, ∣μ0⟩γ0|\mu_0\rangle_{\gamma_0} denotes the known initial coherent photon probe state with polarization γ0\gamma_0, ∣μ⟩γ|\mu\rangle_{\gamma} denotes the unresolved final photon state with polarization γ\gamma, and U(tf,ti)U(t_f,t_i) is the time evolution operator of the entire system. The quantity Aσ1σ2μ\mathcal{A}^{\mu}_{\sigma_1\sigma_2} is an operator acting on the low-energy electronic sector.

In the experiment, the only known information about the final state is the momenta p1\mathbf{p}_1 and p2\mathbf{p}_2 of the photoelectrons detected. In contrast, the photoelectrons spins σ1\sigma_1 and σ2\sigma_2, the final photon state μ\mu, and the state of the low-energy electrons are not known. Therefore, to obtain the probability density of the photoelectrons momenta, we must trace over these degrees of freedom. For the low-energy electronic sector, this is conveniently expressed in terms of the corresponding density matrix ρc\rho_c. Dividing this probability density by the time interval tf−tit_f-t_i then gives the two-photoelectron coincidence rate:

I(2)(p1,p2)≡1tf−ti∑σ1,σ2∑μTr⁡[ρcAσ1σ2μ†(p1,p2;tf,ti)Aσ1σ2μ(p1,p2;tf,ti)].(32)I^{(2)}(\mathbf{p}_1,\mathbf{p}_2) \equiv\frac{1}{t_f-t_i}\sum_{\sigma_1,\sigma_2}\sum_{\mu}\operatorname{Tr}\left[\rho_c\mathcal{A}^{\mu\dagger}_{\sigma_1\sigma_2}(\mathbf{p}_1,\mathbf{p}_2;t_f,t_i)\mathcal{A}^{\mu}_{\sigma_1\sigma_2}(\mathbf{p}_1,\mathbf{p}_2;t_f,t_i)\right]. \tag*{(32)}

Note that while the expression above is written for a pure coherent initial photon state ∣μ0⟩γ0|\mu_0\rangle_{\gamma_0}, it is straightforward to generalize it using a general photon density matrix.

To determine I(2)I^{(2)}, we therefore need to obtain the time evolution operator U(tf,ti)U(t_f,t_i) of H=H0+H1H = H_0 + H_1, which can be done perturbatively in H1H_1 within the interaction picture. Following previous works [1], the relevant interactions that account for the two-photoelectron emission by a single photon take place via a two-step process:

H1=∑γ,ki,σiΓσ1σ2γ(k1,k2)dk1σ1†ck2σ2aγ+∑ki,σiV(k1,k2;k3,k4)dk1σ1†dk2σ2†dk3σ2ck4σ1+h.c..(33)H_1 = \sum_{\gamma,\mathbf{k}_i,\sigma_i}\Gamma^{\gamma}_{\sigma_1\sigma_2}(\mathbf{k}_1,\mathbf{k}_2)d^{\dagger}_{\mathbf{k}_1\sigma_1}c_{\mathbf{k}_2\sigma_2}a_{\gamma} + \sum_{\mathbf{k}_i,\sigma_i}V(\mathbf{k}_1,\mathbf{k}_2;\mathbf{k}_3,\mathbf{k}_4)d^{\dagger}_{\mathbf{k}_1\sigma_1}d^{\dagger}_{\mathbf{k}_2\sigma_2}d_{\mathbf{k}_3\sigma_2}c_{\mathbf{k}_4\sigma_1}+\mathrm{h.c.}. \tag*{(33)}

Here, the first term describes the photoelectron excited by the incoming photon, and Γσσ′γ(p,k)\Gamma^{\gamma}_{\sigma\sigma'}(\mathbf{p},\mathbf{k}) denotes the vertex for this process. For simplicity, we dropped the additional internal degree of freedom α\alpha of the low-energy electrons.

Note that the selection rules and momentum conservation are implicitly satisfied by the vertex function. The second term, on the other hand, denotes the EELS-like process by which a high-energy electron interacts with a low-energy electron to simultaneously create two high-energy electrons. The relevant interaction V(k1,k2;k3,k4)V(\mathbf{k}_1,\mathbf{k}_2;\mathbf{k}_3,\mathbf{k}_4) implicitly satisfies momentum conservation and is left unspecified here. Previously, it has been modeled as a screened Coulomb interaction [1].

For simplicity, we replace the two-step process described in Eq. (33) by an effective single-step process. Moreover, since in the experiment a signal is only observed when the system is in the superconducting state, we can restrict our analysis to the case where the two electrons involved in the two-photoelectron emission process form a Cooper pair singlet with zero center-of-mass momentum. We therefore can write:

H1=H1,c=12∑k,qVeff(k,q)dq↑†d−q↓†c−k↓ck↑aγ+h.c.(34)H_1 = H_{1,c} = \frac{1}{2}\sum_{\mathbf{k},\mathbf{q}} \mathcal{V}_{\mathrm{eff}}(\mathbf{k},\mathbf{q})d_{\mathbf{q}\uparrow}^{\dagger}d_{-\mathbf{q}\downarrow}^{\dagger}c_{-\mathbf{k}\downarrow}c_{\mathbf{k}\uparrow}a_{\gamma}+\mathrm{h.c.} \tag*{(34)}

with an effective “interaction” Veff(k,q)\mathcal{V}_{\mathrm{eff}}(\mathbf{k},\mathbf{q}). Evaluating the time evolution operator U(tf,ti)U(t_f,t_i) perturbatively is straightforward, yielding:

UI(tf,ti)≈1−i∫titfdt H1,I(t),(35)U_I(t_f,t_i) \approx1-i\int_{t_i}^{t_f}dt\,H_{1,I}(t), \tag*{(35)}

where the subscript II denotes the interaction picture (we set ℏ=1\hbar=1). The final expression for I(2)(p1,p2)I^{(2)}(\mathbf{p}_1,\mathbf{p}_2) can be expressed in terms of the gap function, similarly to Ref. [1, 12]. However, the dependence of I(2)(p1,p2)I^{(2)}(\mathbf{p}_1,\mathbf{p}_2) on the energy difference between the two photoelectrons, E−≡E1−E2=p122m−p222mE_- \equiv E_1-E_2 = \frac{p_1^2}{2m}-\frac{p_2^2}{2m}, can be inferred without relying on the final expression. The key point is that, in the process described in Eq. (34) (or in (33)), the two high-energy electrons are created at the same time. Zero center-of-mass momentum constrains their internal momenta to be opposite, q\mathbf{q} and −q-\mathbf{q}, but does not require the two detected asymptotic photoelectrons to have equal energies. Because each photoelectron contributes an outgoing Fourier phase, the fact that the two electrons are created at the same time implies that the two Fourier phases combine as:

eiE1teiE2t=eiE+t,(36)e^{iE_1t}e^{iE_2t}=e^{iE_+t}, \tag*{(36)}

with E+=E1+E2E_+ = E_1+E_2 (here, the energy offset associated with the work function is left implicit). Thus, for the single-photon two-photoelectron process described by Eq. (34), the simultaneous production of the pair of high-energy electrons and the absence of subsequent interactions between them and the material do not introduce an intrinsic relative-time Fourier variable conjugate to the energy difference E−E_-. Of course, due to the momentum dependence in the effective interaction Veff(k,q)\mathcal{V}_{\mathrm{eff}}(\mathbf{k},\mathbf{q}), I(2)(E+,E−)I^{(2)}(E_+,E_-) can depend on E−E_- parametrically through matrix elements and kinematic constraints. However, this dependence does not yield an intrinsic broadening of I(2)(E+,E−)I^{(2)}(E_+,E_-) with respect to E−E_-. Moreover, as explained in the main text, because the observed E−E_- broadening increases as the temperature is decreased, it cannot be attributed to simple quasi-particle lifetime or instrumental resolution. Instead, within the single-photon two-photoelectron channel considered here, an intrinsic E−E_- broadening indicates that the high-energy electrons experiences additional interactions in the material following their production, signaling a breakdown of the so-called sudden approximation.

We thus need to go beyond the sudden approximation and show that I(2)(E+,E−)I^{(2)}(E_+,E_-) acquires a broadening as a function of E−E_-. Going beyond the sudden approximation requires allowing the high-energy electrons to interact with the material before the photoelectron detection. As discussed in [11], the high-energy electrons can interact with any degrees of freedom of the material. The second term in Eq. (33) is an example of such a process, in which the high-energy electrons interact with the low-energy electrons. To generate an intrinsic relative-time variable in the single-photon two-photoelectron channel considered here, one of the outgoing high-energy electrons must interact with the material after they are generated. To show that the resulting E−E_- broadening is independent of the microscopic nature of this additional channel, we describe it by an unspecified bosonic field Φ\Phi that couples to the high-energy electronic density through a number-conserving vertex. Therefore, the H1H_1 Hamiltonian acquires a new term:

H1=H1,c+H1,Φ=12∑k,qVeff(k,q)dq↑†d−q↓†c−k↓ck↑aγ+∑σ,q,QΛ(q,Q)dqσ†dqσΦQ+h.c.,(37)H_1 = H_{1,c}+H_{1,\Phi} = \frac{1}{2}\sum_{\mathbf{k},\mathbf{q}}\mathcal{V}_{\mathrm{eff}}(\mathbf{k},\mathbf{q})d_{\mathbf{q}\uparrow}^{\dagger}d_{-\mathbf{q}\downarrow}^{\dagger}c_{-\mathbf{k}\downarrow}c_{\mathbf{k}\uparrow}a_{\gamma}+\sum_{\sigma,\mathbf{q},\mathbf{Q}}\Lambda(\mathbf{q},\mathbf{Q})d_{\mathbf{q}\sigma}^{\dagger}d_{\mathbf{q}\sigma}\Phi_{\mathbf{Q}}+\mathrm{h.c.}, \tag*{(37)}

where Λ(q,Q)\Lambda(q,Q) is the relevant coupling constant. In order to evaluate I(2)I^{(2)} in Eq. (32), we must now also trace over the bosonic degree of freedom. To proceed, we assume that the collective mode is fluctuating, with ⟨ΦQ⟩Φ=0\langle\Phi_Q\rangle_\Phi=0 and spectral function defined by:

⟨ΦQ1†(t′)ΦQ2(t)⟩Φ=δQ1,Q2∫−∞∞dω2πe−iω(t−t′)SΦ(Q2,ω).(38)\left\langle\Phi_{Q_1}^{\dagger}(t')\Phi_{Q_2}(t)\right\rangle_\Phi=\delta_{Q_1,Q_2}\int_{-\infty}^{\infty}\frac{d\omega}{2\pi}e^{-i\omega(t-t')}S_\Phi(Q_2,\omega). \tag*{(38)}

We can now apply again perturbation theory to obtain the new time evolution operator U(tf,ti)U(t_f,t_i) from H1=H1,c+H1,ΦH_1=H_{1,c}+H_{1,\Phi}. While several terms appear, the leading term describing a later scattering event involving one of the two high-energy electrons is the second-order cross term:

(−i)2∫titfdt∫ttfdt′ H1,Φ,I(t′)H1,c,I(t).(39)(-i)^2\int_{t_i}^{t_f}dt\int_t^{t_f}dt'\,H_{1,\Phi,I}(t')H_{1,c,I}(t). \tag*{(39)}

The reverse ordering term H1,c,I(t′)H1,Φ,I(t)H_{1,c,I}(t')H_{1,\Phi,I}(t) does not contribute in the initial dd-electron vacuum because H1,ΦH_{1,\Phi} conserves the number of high-energy electrons and therefore cannot act before H1,cH_{1,c} creates the two high-energy electrons. Inserting this term in the expression of the transition operator Aσ1σ2μ(p1,p2;tf,ti)\mathcal{A}^{\mu}_{\sigma_1\sigma_2}(\mathbf{p}_1,\mathbf{p}_2;t_f,t_i) in Eq. (31) generates a transition amplitude that is linear in Φ\Phi and of order VeffΛV_{\mathrm{eff}}\Lambda. Since ⟨Φ⟩Φ=0\langle\Phi\rangle_\Phi=0, its interference with the amplitude containing no H1,ΦH_{1,\Phi} term vanishes. The leading non-zero contribution is therefore the product of the linear-in-Φ\Phi amplitude and its Hermitian conjugate, which is of order Veff2Λ2V_{\mathrm{eff}}^2\Lambda^2 and given by:

I~(2)(E+,E−)=∑QSΦ(Q,E+−Ω)∣∑k,qFkC(k,q;Q,p1,p2)Dq(E+,E−;E+−Ω)∣2.(40)\widetilde{I}^{(2)}(E_+,E_-)=\sum_Q S_\Phi(Q,E_+-\Omega)\left|\sum_{k,q}F_k C(k,q;Q,p_1,p_2)D_q(E_+,E_-;E_+-\Omega)\right|^2. \tag*{(40)}

Here, I~(2)\widetilde{I}^{(2)} is the positive contribution with one explicit Φ\Phi vertex in the amplitude and one in its conjugate. We note that this is not the complete coincidence rate at order Λ2\Lambda^2, which also contains the interference of the no-Φ\Phi amplitude with amplitudes containing two explicit Φ\Phi vertices. In Eq. (40), Ω\Omega is the incoming photon frequency, C(k,q;Q,p1,p2)C(k,q;Q,p_1,p_2) is a form factor combining the interaction vertices and the outgoing-state overlaps, and FkF_k is the equal-time anomalous pair amplitude (computed here within mean field theory),

Fk≡⟨c−k↓ck↑⟩c=Δk2ξk2+∣Δk∣2tanh⁡(βξk2+∣Δk∣22)(41)F_k\equiv\langle c_{-k\downarrow}c_{k\uparrow}\rangle_c=\frac{\Delta_k}{2\sqrt{\xi_k^2+|\Delta_k|^2}}\tanh\left(\frac{\beta\sqrt{\xi_k^2+|\Delta_k|^2}}{2}\right) \tag*{(41)}

where Δk\Delta_k is the gap function and ξk\xi_k is the normal-state dispersion of the low-energy electrons. Under the additional assumption that the matrix elements and high-energy-electron propagators for the two scattering processes are approximately equal in the measured kinematic window, the scattering kernel is

Dq(E+,E−;ω)=∫0∞dτ e−Γdτ/2e+i(E+/2−ω−Eq)τcos⁡(E−τ2).(42)D_q(E_+,E_-;\omega)=\int_0^\infty d\tau\,e^{-\Gamma_d\tau/2}e^{+i(E_+/2-\omega-E_q)\tau}\cos\left(\frac{E_-\tau}{2}\right). \tag*{(42)}

where EqE_q is the energy of the high-energy electrons offset by the work function. In the equation above, Γd\Gamma_d is a phenomenological effective damping/dephasing rate of the high-energy-electron propagator, which may receive contributions both from scattering off of the Φ\Phi mode and from other unspecified interactions (these contributions are not computed separately here). The central result underlying the E−E_- broadening of I(2)I^{(2)} is encoded in the function Dq(E+,E−;ω)D_q(E_+,E_-;\omega) defined above. It arises because, due to the exchange of energy between the high-energy electrons and the collective bosonic modes, there are two time scales involved. Specifically, at time tt, the two high-energy electrons are created and at a later time t+τt+\tau, one of them interacts with the collective mode. The outgoing Fourier phases for each scattering process multiply, whereas the two amplitudes corresponding to delayed scattering from either outgoing electron are added coherently. The latter can be approximated as:

eiE1τ+eiE2τ=2eiE+τ/2cos⁡(E−τ2),(43)e^{iE_1\tau}+e^{iE_2\tau}=2e^{iE_+\tau/2}\cos\left(\frac{E_-\tau}{2}\right), \tag*{(43)}

which is the same expression that appears in the integrand of Eq. (42). The emergence of an E−E_{-} broadening can be directly seen by evaluating the integral in Eq. (42) at the resonance condition

ωres=E+2−Eq,(44)\omega_{\mathrm{res}} = \frac{E_{+}}{2} - E_{q}, \tag*{(44)}

which, using ω=E+−Ω\omega= E_{+} - \Omega in Eq. (40), corresponds to a resonance condition for the energy of the dd electrons, 2Eq=2Ω−E+2E_{q} = 2\Omega- E_{+}. At resonance,

∣Dq(E+,E−;ωres)∣2=4Γd2(Γd2+E−2)2.(45)\left|D_{q}(E_{+},E_{-};\omega_{\mathrm{res}})\right|^{2} = \frac{4\Gamma_{d}^{2}}{\left(\Gamma_{d}^{2}+E_{-}^{2}\right)^{2}}. \tag*{(45)}

Thus, the broadening is controlled by the effective high-energy-electron damping/dephasing rate Γd\Gamma_{d}, as well as by the spectral weight of the collective mode through SΦS_{\Phi}, as shown by Eq. (40). A broadening that increases as the temperature decreases is therefore indicative of enhanced damping/dephasing of the outgoing high-energy electrons and/or a temperature-dependent reweighting of the contributing scattering processes. It does not, by itself, necessarily establish that the collective mode becomes softer as the temperature is lowered, since such a conclusion would require independent information about the temperature dependence of SΦS_{\Phi}, which is beyond the scope of this work.

References

  1. [1]Mahmood, F., Devereaux, T., Abbamonte, P. & Morr, D. K. Distinguishing finite-momentum superconducting pairing states with two-electron photoemission spectroscopy. Phys. Rev. B 105, 064515 (2022).DOI
  2. [2]Gao, Q. et al. ARPES detection of superconducting gap sign in unconventional superconductors. Nature Communications 15, 4538–9 (2024).DOI
  3. [3]Gao, Q. et al. Selective hybridization between the main band and the superstructure band in the Bi₂Sr₂CaCu₂O₈+δ superconductor. Phys. Rev. B 101, 014513 (2020).DOI
  4. [4]Kordyuk, A. A. et al. Origin of the peak-dip-hump line shape in the superconducting-state (π, 0) photoemission spectra of Bi₂Sr₂CaCu₂O₈+δ. Phys. Rev. Lett. 89, 077003 (2002).DOI
  5. [5]Dynes, R. C., Narayanamurti, V. & Garno, J. P. Direct measurement of quasiparticle-lifetime broadening in a strongly-coupled superconductor. Phys. Rev. Lett. 41, 1509–1512 (1978).DOI
  6. [6]Sun, X. et al. Temperature evolution of energy gap and band structure in the superconducting and pseudogap states of Bi₂Sr₂CaCu₂O₈+δ superconductor revealed by laser-based angle-resolved photoemission spectroscopy. Chinese Physics Letters 35, 017401 (2018).DOI
  7. [7]Vishik, I. M. et al. Doping-dependent nodal Fermi velocity of the high-temperature superconductor Bi₂Sr₂CaCu₂O₈+δ revealed using high-resolution angle-resolved photoemission spectroscopy. Phys. Rev. Lett. 104, 207002 (2010).DOI
  8. [8]Leitner, T. et al. The CoESCA station at BESSY: Auger electron–photoelectron coincidences from surfaces demonstrated for Ag MNN. Journal of Electron Spectroscopy and Related Phenomena 250, 147075 (2021).
  9. [9]Chernick, M. R. Bootstrap methods : a guide for practitioners and researchers. Wiley series in probability and statistics (Wiley-Interscience, Hoboken, N.J.), 2nd ed. edn.
  10. [10]Hyndman, R. J. & Fan, Y. Sample quantiles in statistical packages. The American Statistician 50, 361–365 (1996).DOI
  11. [11]Berakdar, J. Emission of correlated electron pairs following single-photon absorption by solids and surfaces. Phys. Rev. B 58, 9808–9816 (1998).DOI
  12. [12]Kouzakov, K. A. & Berakdar, J. Photoinduced emission of Cooper pairs from superconductors. Phys. Rev. Lett. 91, 257007 (2003).pubmed.ncbi.nlm.nih.gov/14754144

Paper details

vlnc:W_8fuhwxu2
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents