betaFeedback

WorksPhysics and Astronomy

Magnetic susceptibility and superfluid density of inhomogeneous superconducting films: the Tc-mosaic model

Abstract

Whole-sample and mutual-inductance part-sample measurements of the linear magnetic susceptibility, χχ, of superconducting films are routinely used to obtain the London penetration depth, λ(T)λ(T), assuming uniform superconductivity. However, spatially resolved probes show that many films of current interest are inhomogeneous on micrometer scales. Here we ask whether a distribution of local transition temperatures in a fully gapped film can produce an apparent line-nodal, TT-linear λ(T)λ(T). The film is modeled as a random Voronoi mosaic of domains with TcT_c distributed between Tc,min⁡T_{c,\min} and Tc,max⁡T_{c,\max} with s-wave (also d-wave, for comparison) London response inside each domain. We show that detectable screening appears only below the percolation temperature, TpT_p, and the transition broadens with the spread of TcT_c. For s-wave domains, λ(T)λ(T) remains activated for any Tc,min⁡>0T_{c,\min}>0. Only for Tc,min⁡=0T_{c,\min}=0 does it become a power law, Δλ∝TnΔλ\propto T^n with n≈1.7n\approx1.7, but then the onset of screening drops to Tp=Tc,max⁡/2T_p=T_{c,\max}/2. Weighting the TcT_{c} distribution towards Tc→0T_c\to0 (more low-TcT_{c} domains) can bring nn closer to 1 (n≈1.2n\approx1.2), but simultaneously pushes the apparent transition, TpT_p, more than an order of magnitude below Tc,max⁡T_{c,\max}. For d-wave domains, the linear term survives, and its relative slope is enhanced almost fourfold for Tc,min⁡=0T_{c,\min}=0. Hence, a TT-linear λ(T)λ(T) measured in a film with screening onset within a factor of a few of Tc,max⁡T_{c,\max} by whole-sample or mutual-inductance part-sample measurements cannot originate from TcT_c mosaicity in a fully gapped superconductor.

Introduction

The temperature-dependent London penetration depth, λ(T)\lambda(T), is a primary probe of the superconducting gap structure [1, 2]. It is usually obtained from the small-amplitude, vortex-free linear magnetic susceptibility, χ(T)\chi(T). The susceptibility is measured on the whole sample (e.g., using conventional amplitude-domain AC susceptibility [3] and frequency-domain tunnel-diode resonator [4, 5]) or using a mutual inductance technique, where a macroscopic area of the sample is exposed to a localized magnetic field [6, 7, 8]. Importantly, in thin films, the extraction of λ(T)\lambda(T) from χ(T)\chi(T) requires knowing χ(λ)\chi(\lambda) and assumes a homogeneous superconducting state [5, 9].

The extracted λ(T)\lambda(T) is then used to evaluate the superfluid density, ρs=λ(0)2/λ(T)2\rho_s=\lambda(0)^2/\lambda(T)^2 (or proportional to it, the superconducting phase stiffness). In some cases, its low-temperature behavior is distinctly non-activated. For example, a TT-linear λ(T)\lambda(T) has recently been observed in FeTe-based heterostructures [5]. A closely related local scanning SQUID measurement of FeTe films shows a similar behavior inside the regions of stronger superconductivity [10]. Both suggest a line-nodal order parameter. However, the superconducting state in real films, especially those with complex stoichiometry and fabrication protocols (including the above-mentioned FeTe), is often inhomogeneous, as detected by spatially resolved local measurements on scales of a few to tens of micrometers [10, 11, 12, 13, 14].

Here we ask whether the magnetic susceptibility of a spatially inhomogeneous s-wave film can yield a superfluid density consistent with apparent line-nodal behavior. For comparison, we also study what happens if the film is intrinsically d-wave. Does χ(T)\chi(T) retain a sharp feature at the transition to the normal state? How does χ(T)\chi(T) shape change over the full temperature range? What happens with the low-temperature behavior of ρs(T)\rho_s(T)? To answer these questions, we combine sheet-current London electrodynamics with the percolation theory of random two-dimensional media [15, 16, 17]. In that theory, the superfluid density of a superconductor–insulator mixture maps onto the conductance of a conductor–insulator mixture. We computed the macroscopic magnetic susceptibility for different transition-temperature distributions and analyzed the temperature dependence of the superfluid density and London penetration depth.

Model

The TcT_c–mosaic

The film is tiled by a Poisson–Voronoi tessellation [18]. The seeds are placed randomly with areal density 4/(πℓ2)4/(\pi\ell^2), and each point in the film belongs to the domain of the nearest seed. Here ℓ\ell is the diameter of a disc whose area equals the mean area of all mosaic domains, πℓ2/4\pi\ell^2/4. Domains have random sizes and shapes. We used ℓ=20 μm\ell=20\,\mu\mathrm{m} as the characteristic size suggested by the experiment. The domain ii has its own transition temperature Tc,iT_{c,i}, selected independently and uniformly from the interval [Tc,min⁡,Tc,max⁡][T_{c,\min}, T_{c,\max}] (other distributions are considered in § 3 D). All temperatures are normalized by Tc,max⁡T_{c,\max}: t=T/Tc,max⁡t = T/T_{c,\max}, ti=Tc,i/Tc,max⁡t_i = T_{c,i}/T_{c,\max}, tp=Tp/Tc,max⁡t_p = T_p/T_{c,\max} and tmin⁡=Tc,min⁡/Tc,max⁡t_{\min} = T_{c,\min}/T_{c,\max}, tmax⁡=1t_{\max} = 1. Domain boundaries are sharp. Adjacent superconducting domains share one superconducting condensate, and the supercurrent passes freely between them. A normal domain carries no supercurrent and provides no proximity or Josephson coupling across it.

Figure 1(a) shows an example of a mosaic realization with the broadest distribution, tmin⁡=0t_{\min}=0, where the domains are colored by their TcT_c. Panels (b) and (c) show the superconducting domains below, t=0.45t=0.45, and above, t=0.55t=0.55, the percolation threshold, which in this case is tp(tmin⁡=0)=0.5t_p(t_{\min}=0)=0.5.

Image not extracted

Figure 1. (a) An 800 × 800 μm2\mu\mathrm{m}^{2} realization of the mosaic at tmin⁡=0t_{\min}=0 with mean domain diameter ℓ=20 μm\ell=20\,\mu\mathrm{m} colored by the local TcT_{\mathrm{c}}. (b,c) Superconducting domains (ti>tt_i>t) just below and just above the percolation temperature tp(tmin⁡=0)=0.5t_{\mathrm{p}}(t_{\min}=0)=0.5. The largest connected cluster is shown in red. Below tpt_{\mathrm{p}} it spans the box (periodic boundaries, percolation); above tpt_{\mathrm{p}} only finite non-percolating clusters remain.

Local superfluid density

Within domain ii the London penetration depth λi(T)\lambda_i(T) and the Pearl length Λi(T)\Lambda_i(T) are,

1λi2(T)=tiλ12(0)ρ(TTc,i),Λi(T)=2λi2(T)d=Λ1(0)tiρ(T/Tc,i).(1)\frac{1}{\lambda_i^2(T)}=\frac{t_i}{\lambda_1^2(0)}\rho\left(\frac{T}{T_{c,i}}\right), \qquad \Lambda_i(T)=\frac{2\lambda_i^2(T)}{d}=\frac{\Lambda_1(0)}{t_i\rho(T/T_{c,i})}. \tag*{(1)}

For our numerical calculation, the zero-temperature London penetration depth, λ(0)\lambda(0), of a domain with the highest Tc=Tc,max⁡T_c=T_{c,\max} is set to λ1(0)=1 μm\lambda_1(0)=1\ \mu\mathrm{m}, of the order of the values reported for FeTe-based films [5, 10], so that Λ1(0)=2λ12(0)/d=100 μm\Lambda_1(0)=2\lambda_1^2(0)/d=100\ \mu\mathrm{m} for d=20 nmd=20\ \mathrm{nm}. To avoid extra parameters, for an arbitrary domain ii we use λi−2(0)∝σTc,i\lambda_i^{-2}(0)\propto\sigma T_{c,i} with a common normal-state conductivity σ\sigma. This so-called Homes scaling was reported for a variety of superconductors [19] and, in fact, follows from the BCS theory with disorder for a wide range of scattering rates [20].

Since the main question is whether some spatial distributions of TcT_c in an s-wave mosaic film can produce a superfluid density that looks line-nodal (TT-linear variation at t→0t\to0), we use as intrinsic references for the temperature dependence of the order parameter both a fully gapped isotropic s-wave and a line-nodal d-wave for comparison.

The temperature dependence of the normalized superfluid density, ρ(t)=λ2(0)/λ2(T)\rho(t)=\lambda^2(0)/\lambda^2(T), is computed self-consistently in the full temperature range in the weak-coupling clean limit for a cylindrical Fermi surface, with the superconducting order parameter Δφ(T)=Δ(T)η(φ)\Delta_\varphi(T)=\Delta(T)\eta(\varphi), where η=1\eta=1 for s-wave and η=cos⁡2φ\eta=\cos2\varphi for d-wave, so that Δ(T)\Delta(T) is the maximum gap in both cases. At each temperature, the amplitude Δ(T)\Delta(T) is obtained by solving the self-consistency (gap) equation,

ln⁡1t=2πkBT⟨η2⟩φ∑n≥0⟨η2(φ)[1ωn−1ωn2+Δφ2]⟩φ.(2)\ln\frac{1}{t}=\frac{2\pi k_{\mathrm{B}}T}{\langle\eta^2\rangle_\varphi}\sum_{n\ge0}\left\langle\eta^2(\varphi)\left[\frac{1}{\omega_n}-\frac{1}{\sqrt{\omega_n^2+\Delta_\varphi^2}}\right]\right\rangle_\varphi. \tag*{(2)}

where ωn=πkBT(2n+1)\omega_n=\pi k_{\mathrm{B}}T(2n+1) are the Matsubara frequencies and t=T/Tct=T/T_c. The superfluid density follows from the Matsubara sum,

ρ(t)=2πkBT∑n≥0⟨Δφ2(ωn2+Δφ2)3/2⟩φ.(3)\rho(t)=2\pi k_{\mathrm{B}}T\sum_{n\ge0}\left\langle\frac{\Delta_\varphi^2}{\left(\omega_n^2+\Delta_\varphi^2\right)^{3/2}}\right\rangle_\varphi. \tag*{(3)}

There are no adjustable parameters: TcT_c fixes Δ(T)\Delta(T) and ρ(t)\rho(t).

For an s-wave order parameter, Δ(0)=1.764 kBTc\Delta(0)=1.764\,k_{\mathrm{B}}T_c, and ρ≃1−2πΔ(0)/kBTe−Δ(0)/kBT\rho\simeq1-\sqrt{2\pi\Delta(0)/k_{\mathrm{B}}T}e^{-\Delta(0)/k_{\mathrm{B}}T} is exponentially small at low temperatures. The dirty-limit s-wave form, ρ=(Δ(T)/Δ(0))tanh⁡(Δ(T)/2kBT)\rho=(\Delta(T)/\Delta(0))\tanh(\Delta(T)/2k_{\mathrm{B}}T), would differ from Eq. (3) by at most 0.7%0.7\% at t≃0.7t\simeq0.7, and it has the same exponential low-temperature behavior [21].

For a d-wave order parameter, Δ(0)=2.14 kBTc\Delta(0)=2.14\,k_{\mathrm{B}}T_c, and ρ≃1−(2ln⁡2)kBT/Δ(0)=1−0.6487T/Tc\rho\simeq1-(2\ln2)k_{\mathrm{B}}T/\Delta(0)=1-0.6487T/T_c at low temperature, i.e., Δλ/λ(0)=0.324 T/Tc\Delta\lambda/\lambda(0)=0.324\,T/T_c [1]. The impurity rounding of the d-wave linear term, T→T2/(T+T∗)T\to T^2/(T+T^*), is not included. This formula was derived for strong (unitary) limit scattering to explain the experiments in high-TcT_c cuprates [22]. In a more common weak-scattering (Born) limit, the T→T2T\to T^2 rounding becomes noticeable only above a very large scattering rate, Γ≈0.17\Gamma\approx0.17, where Γ=ℏ/(2πkBTc0τ)\Gamma=\hbar/(2\pi k_{\mathrm{B}}T_{c0}\tau), τ\tau is the total (transport plus spin-flip) Born scattering time, and Tc0T_{c0} is the clean-limit TcT_c, when the transition temperature is almost half that of the clean state, Tc(Γ=0.17)=0.547Tc(Γ=0)T_c(\Gamma=0.17)=0.547T_c(\Gamma=0). For reference, the critical value at which d-wave superconductivity is quenched by non-spin-flip scattering is Tc(Γcrit=0.28)=0T_c(\Gamma_{\mathrm{crit}}=0.28)=0 [23].

Each domain enters the calculation only through what we call its “contrast” relative to the strongest domain,

ci(T)≡Λi(T)Λ1(T)={ρ(t)tiρ(t/ti),t<ti,∞,t≥ti.(4)c_i(T)\equiv\frac{\Lambda_i(T)}{\Lambda_1(T)}= \begin{cases} \frac{\rho(t)}{t_i\rho(t/t_i)}, & t<t_i,\\[6pt] \infty, & t\ge t_i. \end{cases} \tag*{(4)}

Sheet currents in the TcT_c-mosaic and homogenization

Even in an ultra-thin film in a perpendicular magnetic field, the supercurrent density is far from uniform throughout the thickness [9]. However, for the magnetic fields outside the film, and therefore for the magnetic moment, mm, and magnetic susceptibility, χ=m/(HV)\chi=m/(HV) (VV is the sample volume), only the thickness-integrated sheet current, II (current per unit width), matters. With Pearl length Λ=2λ2/d\Lambda=2\lambda^2/d, this 2D sheet description reproduces full three-dimensional solutions for λ≳4d\lambda\gtrsim4d and d/2a≲10−4d/2a\lesssim10^{-4} [9].

The London equation of the sheet reads,

I=−2μ0Λ(A−Φ02π∇θ)=2meμ0eΛvs,(5)I=-\frac{2}{\mu_0\Lambda}\left(\mathbf{A}-\frac{\Phi_0}{2\pi}\nabla\theta\right)=\frac{2m_e}{\mu_0 e\Lambda}\mathbf{v}_s, \tag*{(5)}

where mem_e and ee are the electron mass and charge, Φ0\Phi_0 is the flux quantum. Here, θ\theta is the phase of the order parameter, and vs\mathbf{v}_s is the superfluid velocity, both averaged over the thickness. The kinetic energy per unit area is μ0ΛI2/4\mu_{0}\Lambda I^{2}/4. Since II is the actual current and ρs∝1/λ2∝1/Λ\rho_{\mathrm{s}}\propto1/\lambda^{2}\propto1/\Lambda, Eq. (5) can be thought of as Ohm’s law with ρs\rho_{\mathrm{s}} as the scalar “conductivity” and vs\mathbf{v}_{\mathrm{s}} as the vector “electric field”. A random network of domains with different ρs\rho_{\mathrm{s}} is therefore the superconducting analog of a random resistor network.

The problem has three well-separated length scales: the microscopic ones (ξi\xi_i, λi\lambda_i, dd), the mesoscopic domain size ℓ=20 μm\ell=20\,\mu\mathrm{m}, and the macroscopic sample lateral size a=3 mma=3\,\mathrm{mm}. The coherence length, ξi\xi_i, plays no role, since we solve the London equation. In the calculations, we make the following assumptions:

  1. Local London response. ξ≪ℓ\xi\ll\ell, so each domain obeys Eq. (5) with its own Λi(T)\Lambda_i(T). Domain walls are sharp on the scale ℓ\ell. The proximity length in a normal domain, ∼ℏD/2πkBT≲0.1 μm\sim\sqrt{\hbar D/2\pi k_{\mathrm{B}}T}\lesssim0.1\,\mu\mathrm{m}, is negligible.

  1. Thin-sheet limit. d≪ℓd\ll\ell, and the sheet conditions are met: λ≥1 μm=50d\lambda\geq1\,\mu\mathrm{m}=50d and d/2a=3.3×10−6d/2a=3.3\times10^{-6}. Only the thickness-integrated current enters, both in the local physics and in the magnetic fields outside the film.

  1. Vortex-free linear response. The excitation magnetic field amplitude is small and no vortices enter. The fluxoid (flux plus the contribution of screening currents) around every hole in the superconducting network, i.e., every normal island or cluster, keeps its zero-field value.

  1. Scale separation. ℓ≪a\ell\ll a. Locally, the macroscopic Meissner current of the disc looks like a uniform mean current Iˉ\bar{\mathbf{I}}. On the scale ℓ\ell the TcT_{\mathrm{c}} inhomogeneity only redistributes this current around the domains.

  1. Stray fields of the redistributed current. The non-uniform part of I\mathbf{I} produces magnetic fields that decay away from the film over a distance ∼ℓ\sim\ell. Their energy is kept in the local problem but does not affect the macroscopic field.

Because ∇⋅I=0\nabla\cdot\mathbf{I}=0, the current can be written with a stream function, g(r)g(\mathbf{r}), I=Iˉ+∇×(gz^)\mathbf{I}=\bar{\mathbf{I}}+\nabla\times(g\hat{\mathbf{z}}). Function g(r)g(\mathbf{r}) obeys Born–von Kármán periodic boundary conditions on the 2D L×LL\times L computational box. The mean current is carried by the non-periodic linear part of the stream function. The energy per unit area of the local problem is

4μ0E[g]=⟨Λ(r)∣Iˉ+∇×gz^∣2⟩+∑k≠0∣k∣∣gk∣2,(6)\frac{4}{\mu_{0}}E[g]=\left\langle\Lambda(\mathbf{r})\left|\bar{\mathbf{I}}+\nabla\times g\hat{\mathbf{z}}\right|^{2}\right\rangle+\sum_{\mathbf{k}\ne0}|\mathbf{k}||g_{\mathbf{k}}|^{2}, \tag*{(6)}

where gkg_{\mathbf{k}} are the Fourier amplitudes of g(r)g(\mathbf{r}). The first term is the kinetic energy. The second is the energy of the magnetic field produced by the current pattern in the half-spaces above and below the film (∣z∣>d/2|z|>d/2). The field energy inside the film is negligible for d≪ℓd\ll\ell. For a uniform sheet, the two terms give ∣k∣(1+Λ∣k∣)∣gk∣2|\mathbf{k}|(1+\Lambda|\mathbf{k}|)|g_{\mathbf{k}}|^{2}, which is Pearl’s kernel [24]. The minimum of Eq. (6) at fixed Iˉ\bar{\mathbf{I}} defines the homogenized Pearl length, Λeff\Lambda_{\mathrm{eff}},

min⁡gE=μ04ΛeffIˉ2,G≡ΛeffΛ1(T),(7)\min_{g} E=\frac{\mu_{0}}{4}\Lambda_{\mathrm{eff}}\bar{I}^{2},\qquad G\equiv\frac{\Lambda_{\mathrm{eff}}}{\Lambda_{1}(T)}, \tag*{(7)}

where Λ1(T)\Lambda_{1}(T) is the Pearl length of a “strongest superconductivity” domain with Tc=Tc,maxT_{\mathrm{c}}=T_{\mathrm{c,max}}, Eq. (1) at ti=1t_i=1. The local Pearl length in Eq. (6) can be written as Λ(r)=Λ1(T)c(r)\Lambda(\mathbf{r})=\Lambda_{1}(T)c(\mathbf{r}), with what we call a “contrast” c(r)=cic(\mathbf{r})=c_i for r\mathbf{r} in domain ii Eq. (4). The dimensionless ratio is G=1G=1 for a uniform film with Tc=Tc,maxT_{\mathrm{c}}=T_{\mathrm{c,max}} and G≥1G\geq1 otherwise.

The Euler–Lagrange equation of Eq. (6) is the London equation for the sheet current I\mathbf{I}, with Pearl’s Λ\Lambda as the natural length scale,

μ02z^⋅∇×[Λ(r)I]=−Bz(r),(8)\frac{\mu_0}{2}\hat{\mathbf{z}}\cdot\nabla\times[\Lambda(\mathbf{r})\mathbf{I}]=-B_z(\mathbf{r}), \tag*{(8)}

where Bz(r)B_z(\mathbf{r}) is the magnetic field produced by the current I\mathbf{I} in the plane of the film. Inside a normal domain, Λ→∞\Lambda\to\infty forces I→0\mathbf{I}\to0, so the stream function, including its linear part, is constant there. This constant is a free variable of the minimization of Eq. (6). Stationarity with respect to this constant sets the change of the fluxoid around the domain to zero, which is assumption 3.

The magnetic term describes the screening of the redistributed currents by their own stray fields. It is small when ℓ≪Λeff\ell\ll\Lambda_{\mathrm{eff}}. Without it, Eq. (6) is the variational principle for the effective conductance of a random conductor with local conductivity ρ(r)∝1/Λ(r)\rho(\mathbf{r})\propto1/\Lambda(\mathbf{r}) [15]. Two exact properties of this problem are used below.

  • For a statistically symmetric two-phase random mixture in two dimensions, the Keller–Dykhne duality gives ρeff=ρaρb\rho_{\mathrm{eff}}=\sqrt{\rho_a\rho_b} [25, 26].

  • Because ρeff\rho_{\mathrm{eff}} is a homogeneous first-degree function of the local superfluid densities and the current pattern minimizes EE, a small change δρ(r)\delta\rho(\mathbf{r}) changes ρeff\rho_{\mathrm{eff}} by

δρeff=⟨δρ(r)u2(r)⟩,u(r)=∣vs(r)∣∣⟨vs⟩∣.(9)\delta\rho_{\mathrm{eff}}=\left\langle\delta\rho(\mathbf{r})u^2(\mathbf{r})\right\rangle,\qquad u(\mathbf{r})=\frac{|\mathbf{v}_s(\mathbf{r})|}{|\langle\mathbf{v}_s\rangle|}. \tag*{(9)}

Here u(r)u(\mathbf{r}) is the magnitude of the local superfluid velocity, vs∝ΛI\mathbf{v}_s\propto\Lambda\mathbf{I}, Eq. (5), normalized by the magnitude of its average over the film. In the kinetic limit, u=Λ(r)∣I(r)∣/(ΛeffIˉ)u=\Lambda(\mathbf{r})|\mathbf{I}(\mathbf{r})|/(\Lambda_{\mathrm{eff}}\bar{I}). As in a resistor network, where a poorly conducting inclusion carries little current but a large electric field, a weak domain carries a smaller current but a larger supervelocity: u<1u<1 in strong domains and u>1u>1 in weak ones. Normal domains do not contribute to Eq. (9), since δρ=0\delta\rho=0 there.

Total magnetic susceptibility of a thin disc

Once Λeff(T)\Lambda_{\mathrm{eff}}(T) is known, the film is a uniform sheet with Pearl length Λeff\Lambda_{\mathrm{eff}}. It was shown previously that in the sheet-current regime, the total magnetic susceptibility, χ\chi, of a disc of radius aa for arbitrary λ\lambda and aspect ratios d/2a≤10−4d/2a\leq10^{-4}, is [9]

∣χ∣χ0=F(Λeffa),χ0=8a3πd,(10)\frac{|\chi|}{\chi_0}=F\left(\frac{\Lambda_{\mathrm{eff}}}{a}\right),\qquad\chi_0=\frac{8a}{3\pi d}, \tag*{(10)}

with χ0\chi_0 the ideal (Λ→0)(\Lambda\to0) thin-disc value [27]. The sheet function is (with ϵ=Λeff/a\epsilon=\Lambda_{\mathrm{eff}}/a),

1F(ϵ)−1=ϵ[3πln⁡(1+c2ϵ)+323π+c31+c4ϵ+c5ϵ(1+c6ϵ)2],(11)\frac{1}{F(\epsilon)}-1=\epsilon\left[\frac{3}{\pi}\ln\left(1+\frac{c_2}{\epsilon}\right)+\frac{32}{3\pi}+\frac{c_3}{1+c_4\epsilon}+\frac{c_5\epsilon}{(1+c_6\epsilon)^2}\right], \tag*{(11)}

with c2=0.0401744c_2=0.0401744, c3=0.8898535c_3=0.8898535, c4=8.093975c_4=8.093975, c5=20.0621225c_5=20.0621225 and c6=94.44745c_6=94.44745 (accuracy 0.02%) [9]. Its limits are

  • 1−F≃(3/π)ϵln⁡(3.57/ϵ)1-F\simeq(3/\pi)\epsilon\ln(3.57/\epsilon) for ϵ→0\epsilon\to0;

  • F≃(3π/32)ϵ−1F\simeq(3\pi/32)\epsilon^{-1} for ϵ≫1\epsilon\gg1.

For a=3a=3 mm the crossover Λeff=a\Lambda_{\mathrm{eff}}=a corresponds to ρeff=Λ1(0)/a=0.033\rho_{\mathrm{eff}}=\Lambda_1(0)/a=0.033. Over most of the temperature range the disc is therefore in the strong-screening regime, Λeff≪a\Lambda_{\mathrm{eff}}\ll a, where ∣χ∣|\chi| is close to χ0\chi_0 and the deficit is small, 1−∣χ∣/χ0≃(3/π)(Λeff/a)ln⁡(3.57a/Λeff)1-|\chi|/\chi_0\simeq(3/\pi)(\Lambda_{\mathrm{eff}}/a)\ln(3.57a/\Lambda_{\mathrm{eff}}), i.e. nearly linear in Λeff∝1/ρeff\Lambda_{\mathrm{eff}}\propto1/\rho_{\mathrm{eff}}, with a logarithmic correction.

Numerical Procedure

The effective Pearl length Λeff(T)\Lambda_{\mathrm{eff}}(T) was computed in six steps. The parameters used are listed in Table I.

ParameterValue
disc radius, thicknessa=3a=3 mm, d=20d=20 nm
λ1(0)\lambda_1(0), Λ1(0)\Lambda_1(0)11 μm, 100100 μm
domain diameter ℓ\ell2020 μm, Voronoi
local TcT_cuniform in [tmin⁡,1]Tc,max⁡[t_{\min},1]T_{c,\max}
λi−2(0)\lambda_i^{-2}(0)∝Tc,i\propto T_{c,i}
box, grid800800 μm, 2562256^2 (h=3.125h=3.125 μm)
realizations3 (two current directions each)
normal-domain contrastc=106c=10^6
CG tolerance10−1010^{-10} (relative residual)

Table I. Parameters used in the calculation.

Step 1: TcT_c-mosaic. A periodic square box of side L=800L=800 μm (40 domain diameters) is filled with a Poisson number of seeds, with mean L2/(πℓ2/4)≃2040L^2/(\pi\ell^2/4)\simeq2040. The box is discretized by a square N×NN\times N grid with N=256N=256, so the grid spacing (the side of one grid cell) is h=L/N=3.125h=L/N=3.125 μm, and a domain of mean diameter spans 6.4 grid cells. Each grid node (ih,jh)(ih,jh) and each edge midpoint, ((i+12)h,jh)((i+\frac{1}{2})h,jh) and (ih,(j+12)h)(ih,(j+\frac{1}{2})h), is assigned to the domain of its nearest seed, with distances measured periodically across the box edges.

Step 2: contrast and local superfluid density. At each temperature, contrast ci(t)c_i(t) is evaluated from Eq. (4). The function ρ(t)\rho(t) of each pairing state is tabulated once on a fine grid by the same code. At each tabulated temperature, Eq. (2) is solved for Δ\Delta by root bracketing, with the Matsubara sum truncated at 4000 terms plus its analytic tail. Equation (3) is then summed up to ωn≃40Δ\omega_n\simeq40\Delta (at least 2000 terms), again with the analytic tail. The angular average uses 128 points on [0,π/2][0,\pi/2] for d-wave and is trivial for s-wave. Below t=0.02t=0.02 the angular grid cannot resolve the d-wave nodes, so the exact limit shown previously, 1−0.648 t1-0.648\,t, is used. For s-wave, ρ=1\rho=1 to machine precision below t=0.01t=0.01. Normal domains get c=106c=10^6, which makes their current negligible (below 10−610^{-6} of the mean).

Step 3: discretization. The stream function gg is defined on the grid nodes. With periodic forward differences DxD_x, DyD_y, the current components Ix=Iˉx+DygI_x=\bar{I}_x+D_y g and Iy=Iˉy−DxgI_y=\bar{I}_y-D_x g live on the edges, where the corresponding contrasts c(y)c^{(y)} and c(x)c^{(x)} are evaluated. This staggered (Marker-and-Cell, MAC-type) grid conserves current exactly cell by cell. The kinetic energy is Λ1⟨c(y)⟩Ix2+⟨c(x)⟩Iy2\Lambda_1\langle c^{(y)}\rangle I_x^2+\langle c^{(x)}\rangle I_y^2. The magnetic term is evaluated by fast Fourier transform with the lattice wavenumber ∣k∣h=(2/h)[sin⁡2(kxh/2)+sin⁡2(kyh/2)]1/2|\mathbf{k}|_h=(2/h)[\sin^2(k_xh/2)+\sin^2(k_yh/2)]^{1/2}.

Step 4: linear system. Minimizing Eq. (6) gives

[Λ1L+M]g=b.[\Lambda_1L+M]g=b.

Here L=DxTC(x)Dx+DyTC(y)DyL=D_x^{\mathsf{T}}C^{(x)}D_x+D_y^{\mathsf{T}}C^{(y)}D_y is the kinetic (weighted-Laplacian) operator and MM is the magnetic operator, diagonal in k\mathbf{k}. The right-hand side is b=−Λ1DxTC(y)b=-\Lambda_1D_x^{\mathsf{T}}C^{(y)} for Iˉ=x^\bar{\mathbf{I}}=\hat{x} and b=+Λ1DxTC(x)b=+\Lambda_1D_x^{\mathsf{T}}C^{(x)} for Iˉ=y^\bar{\mathbf{I}}=\hat{y}. One node is fixed to remove the constant.

When normal domains are present, the edge contrasts span six orders of magnitude, and conjugate gradients [28] with a diagonal preconditioner fail to converge. For example, for tmin⁡=0t_{\min}=0 at t=0.4t=0.4 they do not reach a required relative residual of 10−1010^{-10} in 2×1042\times10^4 iterations. We therefore factorize Λ1L\Lambda_1L exactly by sparse LU [29]. The elimination order controls the fill-in of the factors. With column approximate minimum degree (COLAMD) ordering [30], the factors contain 1.2×1071.2\times10^7 nonzeros (37 times as many as LL) and take about 1 s to compute on one CPU core, against 6.7×1076.7\times10^7 non-zeros and 15 s in natural order.

Step 5: superfluid density. The energy of the solution gives GxG_x and GyG_y, through Eq. (7). The values of 1/G1/G for three realizations and two directions are averaged, and the standard error of the six values is recorded,

ρeff(t)=ρ(t)⟨G−1⟩.(12)\rho_{\mathrm{eff}}(t)=\rho(t)\left\langle G^{-1}\right\rangle. \tag*{(12)}

The kinetic-only value ρkin\rho_{\mathrm{kin}}, obtained from the LU solution before iteration, measures the stray-field correction.

Step 6: magnetic susceptibility. Λeff=Λ1(0)/ρeff\Lambda_{\mathrm{eff}}=\Lambda_1(0)/\rho_{\mathrm{eff}} is inserted into Eqs. (10) and (11) with a=3a=3 mm.

Temperature grid. Each distribution (tmin⁡=0.9t_{\min}=0.9, 0.7, 0.5, 0.2 and 0) and pairing state uses 144 temperatures: 100 uniformly spaced in [0,0.995][0,0.995], 12 logarithmically spaced in [0.005,0.1][0.005,0.1], 30 concentrated around the percolation temperature, and two at t=0.997t=0.997 and 0.999. For the uniform film, tmin⁡=1t_{\min}=1, the points at t≥1t\geq1 are dropped, leaving 136.

The numerical calculations were validated using several criteria. The result expected for a uniform film, G=1G=1, was reproduced to machine precision. The Keller–Dykhne duality was checked for a 50/50 two-phase Voronoi mosaic with contrast 10 (kinetic only), where the solver gives G=3.142G=3.142 (average over three realizations and two directions), against the expected 10=3.162\sqrt{10}=3.162. For contrast 100, it gives 10.03 against 10. For tmin⁡=0t_{\min}=0, grid refinement to 5122512^2 changes GG by less than 0.3% away from tpt_p, and doubling the box to 1600 μ\mum (≈ 8000 domains) leaves ρeff\rho_{\mathrm{eff}} unchanged within the realization scatter for t≤0.45t\leq0.45 and sharpens the approach to tpt_p. Including the magnetic term changes ρeff\rho_{\mathrm{eff}} by at most 1.3% at t=0t=0 and by at most 6% close to tpt_p, where normal clusters are large.

Results

The effective superfluid density

The domains that are superconducting at temperature tt occupy a random fraction x(t)=(1−t)/(1−tmin⁡)x(t)=(1-t)/(1-t_{\min}) of the film. Site percolation on the Poisson–Voronoi tessellation has a threshold of exactly 1/21/2 [31]. A macroscopic superfluid density therefore exists only below

tp=1+tmin⁡2,(13)t_p=\frac{1+t_{\min}}{2}, \tag*{(13)}

which is the median local TcT_c. Between tpt_p and 1 the film contains only finite superconducting clusters [Fig. 1(c)]. Each cluster screens on its own, but a cluster of size R≪ΛR\ll\Lambda has moment ∼R4H/Λ\sim R^4H/\Lambda. Their total relative to a connected film is of order R2/(aΛ)≲10−3R^2/(a\Lambda)\lesssim10^{-3} for R∼ℓR\sim\ell, which is invisible on the scale of the figures.

Figure 2 summarizes the results. A narrow distribution, tmin⁡=0.9t_{\min}=0.9, gives an almost uniform curve with the onset of screening at tp=0.95t_p=0.95. For broader distributions, the ρeff(t)\rho_{\mathrm{eff}}(t) curves shift to lower temperatures. As far as the low-temperature saturation is concerned, the results for the two pairing symmetries are very different. For s-wave, Figure 2(a), the flat region shrinks, yet remains finite and only disappears for tmin⁡=0t_{\min}=0, as expected. However, for d-wave domains, the low-temperature part is tt-linear from the start. As shown in Figure 2(b), broadening the TcT_c distribution also shifts the curves to lower temperatures, and the effective transition temperature follows the same trend as in the s-wave case, but the low-temperature behavior remains linear.

Image not extracted

Figure 2. Effective (measured) superfluid density ρeff=Λ1(0)/Λeff\rho_{\mathrm{eff}} = \Lambda_1(0)/\Lambda_{\mathrm{eff}} of the mosaic (solid lines) and the area average (see text) of the local superfluid density (dashed lines) for (a) s-wave, and (b) d-wave domains. The label λ(0)=1 μm\lambda(0)=1\,\mu\mathrm{m} denotes λ1(0)\lambda_1(0), the zero-temperature penetration depth of the strongest domains (Tc=Tc,max⁡T_c=T_{c,\max}); the effective λeff(0)=λ1(0)/ρeff(0)\lambda_{\mathrm{eff}}(0)=\lambda_1(0)/\sqrt{\rho_{\mathrm{eff}}(0)} of the film is larger.

For contrast, the dashed lines in Figure 2 show the area average of the local superfluid density, ⟨tiρ(t/ti)⟩\langle t_i\rho(t/t_i)\rangle, which counts all superconducting domains regardless of connectivity and therefore extends to t=1t=1. It is the upper bound on ρeff\rho_{\mathrm{eff}}. Therefore, such a simplified analysis of the inhomogeneous state, even with mosaic domains, does not reflect the experiment and is not correct.

Magnetic susceptibility in the TcT_c-mosaic model

Figure 3(a,Figure 3b) shows ∣χ(t)∣/χ0|\chi(t)|/\chi_0 over the full temperature range for s-wave (a) and d-wave (b) cases, together with its temperature derivative (c,d), respectively. As expected, χ=0\chi=0 above tpt_p, not only above t=1t=1. (Note that finite superconducting clusters retain some diamagnetic response, which is neglected in our homogenized-disc calculation.) Similarly to the superfluid density discussed above, the s-wave curves saturate at low temperatures, whereas the d-wave curves do not. Of course, in a real experiment, the minimum measurement temperature will limit the observation of this behavior. Furthermore, an important feature to compare with the experiment is the sharpness of the superconducting transition. For the uniform film (tmin⁡=1t_{\min}=1), −dχ/dt-d\chi/dt jumps to a finite value at TcT_c, whereas in the mosaic it rises continuously from zero at tpt_p, Figure 3(c,Figure 3d) (see § 4), and the whole curve broadens more and more as tmin⁡t_{\min} decreases.

Image not extracted

Figure 3. (a,b) Magnetic susceptibility of a disc (a=3a = 3 mm, d=20d = 20 nm, λ1(0)=1 μ\lambda_1(0) = 1\,\mum, labeled λ(0)\lambda(0) in the panels) for s-wave and d-wave domains. tmin⁡=1t_{\min}=1: uniform film. Dotted: uniform film with the area-averaged superfluid density [dashed lines in Figure 2], tmin⁡=0.9t_{\min}=0.9 to 0 from top to bottom. (c,d) −d(∣χ∣/χ0)/dt-\mathrm{d}(|\chi|/\chi_0)/\mathrm{d}t, smoothed (except for tmin⁡=1t_{\min}=1) with a Gaussian of width δt=0.01\delta t=0.01 to suppress the steps produced by individual cluster connections in the finite box. A kink in χ(t)\chi(t) would appear as a step.

Low-temperature penetration depth

Figure 4 shows Δλeff(t)=λeff(t)−λeff(0)\Delta\lambda_{\mathrm{eff}}(t)=\lambda_{\mathrm{eff}}(t)-\lambda_{\mathrm{eff}}(0), the quantity a whole-sample measurement reports after calibration [5, 9].

For s-wave domains, Figure 4(a), the behavior remains activated for every tmin⁡>0t_{\min}>0, with an activation energy set by the weakest domains, Δ(0)≈1.76 kBTc,min⁡\Delta(0)\approx1.76\,k_B T_{c,\min}. The amplitude is larger for broader distributions, consistent with the reduction of the superfluid density. For tmin⁡=0t_{\min}=0, domains with ti≲tt_i\lesssim t keep losing their condensate down to t→0t\to0. For this uniform distribution, the area average decreases as Δ⟨ρ⟩∝t2\Delta\langle\rho\rangle\propto t^2: the domains that lose their condensate have ti≲tt_i\lesssim t, and their superfluid density is ∝ti\propto t_i.

Image not extracted

Figure 4. Low-temperature part of Δλeff(t)=λeff(t)−λeff(0)\Delta\lambda_{\mathrm{eff}}(t)=\lambda_{\mathrm{eff}}(t)-\lambda_{\mathrm{eff}}(0), in nm, for (a) s-wave and (b) d-wave domains, with λ1(0)=1 μm\lambda_1(0)=1\,\mu\mathrm{m} (labeled λ(0)\lambda(0) in the panel). For broad distributions, the curves leave the frame as t→tpt\to t_p, where λeff\lambda_{\mathrm{eff}} diverges.

For the film, the decrease is several times larger and the exponent somewhat smaller (≈1.8\approx1.8), because the weak domains carry the largest supervelocity, u>1u>1, and their loss enters Eq. (9) with the weight u2u^2. Power-law fits, Δλeff∝tn\Delta\lambda_{\mathrm{eff}}\propto t^n, over windows within t≤0.3t\leq0.3 give n=1.6n=1.6–1.81.8 for tmin⁡=0t_{\min}=0 and n>4n>4 for tmin⁡=0.2t_{\min}=0.2; the local exponent, dln⁡Δλeff/dln⁡td\ln\Delta\lambda_{\mathrm{eff}}/d\ln t, of the tmin⁡=0t_{\min}=0 curve never falls below 1.5.

For d-wave domains, Figure 4(b), where λi−2(0)∝ti\lambda_i^{-2}(0)\propto t_i, the superfluid density of every domain at low temperature is

tiρd(t/ti)=ti(1−0.648tti)=ti−0.648t.(14)t_i\rho_d(t/t_i)=t_i\left(1-0.648\frac{t}{t_i}\right)=t_i-0.648t. \tag*{(14)}

Every domain therefore loses superfluid density at the same absolute rate, independent of its TcT_c. By Eq. (9), the film loses it at

dρeffdt∣t→0=−0.648⟨u2⟩,(15)\left.\frac{d\rho_{\mathrm{eff}}}{dt}\right|_{t\to0}=-0.648\langle u^2\rangle, \tag*{(15)}

where ⟨u2⟩≥1\langle u^2\rangle\geq1 is the mean-square supervelocity of the t=0t=0 current pattern. For tmin⁡=0t_{\min}=0, ⟨u2⟩=1.48\langle u^2\rangle=1.48. The relative slope ∣dln⁡ρeff/dt∣|d\ln\rho_{\mathrm{eff}}/dt| is then 0.648×1.48/0.385=2.490.648\times1.48/0.385=2.49 instead of 0.648. The initial linear slope of Δλeff/λeff(0)\Delta\lambda_{\mathrm{eff}}/\lambda_{\mathrm{eff}}(0) is therefore enhanced by a factor ⟨u2⟩/ρeff(0)≈3.8\langle u^2\rangle/\rho_{\mathrm{eff}}(0)\approx3.8 relative to a uniform film, while the exponent is unchanged. Because λeff(0)=λ1(0)/ρeff(0)=1.61λ1(0)\lambda_{\mathrm{eff}}(0)=\lambda_1(0)/\sqrt{\rho_{\mathrm{eff}}(0)}=1.61\lambda_1(0), the absolute slope of Δλeff(t)\Delta\lambda_{\mathrm{eff}}(t) in Figure 4(b) is enhanced by ≈6.2\approx6.2 relative to a uniform film with λ(0)=λ1(0)\lambda(0)=\lambda_1(0). The computed initial slope, dρeff/dt=−0.974d\rho_{\mathrm{eff}}/dt=-0.974 between t=0t=0 and 0.005, agrees with the prediction −0.648×1.48=−0.961-0.648\times1.48=-0.961 within the realization scatter of ⟨u2⟩\langle u^2\rangle (±0.05\pm0.05). For tmin⁡=0.5t_{\min}=0.5, ⟨u2⟩=1.020\langle u^2\rangle=1.020. A large linear slope in a whole-sample measurement of an inhomogeneous nodal film therefore does not by itself imply a large nodal density of states. It may simply reflect a reduced effective superfluid density.

Non-uniform TcT_c distributions

The uniform distribution is not special. For an arbitrary normalized distribution of local critical temperatures, P(ti)P(t_i), the area-averaged superfluid density of s-wave domains decreases by

Δ⟨ρ⟩(t)=∫01P(ti)ti[1−ρ(tti)] dti=t∫01P(ti)f(tti) dti,(16)\begin{aligned} \Delta\langle\rho\rangle(t)&=\int_0^1 P(t_i)t_i\left[1-\rho\left(\frac{t}{t_i}\right)\right]\,dt_i\\ &=t\int_0^1 P(t_i)f\left(\frac{t}{t_i}\right)\,dt_i, \tag*{(16)} \end{aligned}

where ρ(x)=0\rho(x)=0 for x≥1x\geq1 and f(x)=[1−ρ(x)]/xf(x)=[1-\rho(x)]/x. The function ff is bounded, 0≤f≤10\leq f\leq1, exponentially small for x→0x\to0, and is equal to 1/x1/x for x≥1x\geq1. Hence, a domain with ti≫tt_i\gg t contributes much less than tt, while the domains with ti≲tt_i\lesssim t, which contribute at most tt each, occupy a fraction of the film that vanishes as t→0t \to0. Therefore, Δ⟨ρ⟩/t→0\Delta\langle\rho\rangle/t \to0 for any distribution of local TcT_c (formally, this follows from Lebesgue’s dominated-convergence theorem: if functions converge pointwise and are all bounded by one integrable function, their integrals converge to the integral of the limit. Here, P(ti)f(t/ti)→0P(t_i)f(t/t_i) \to0 for every ti>0t_i > 0 and is bounded by P(ti)P(t_i)). On cooling, the area-averaged superfluid density always approaches its zero-temperature value with zero slope, d⟨ρ⟩/dt→0\mathrm{d}\langle\rho\rangle/\mathrm{d}t \to0, and has no TT-linear term.

For example, for a general power-law distribution, P(ti)=(1+α)tiαP(t_i)=(1+\alpha)t_i^\alpha on (0,1](0,1] with α>−1\alpha>-1, Eq. (16) gives exactly Δ⟨ρ⟩∝tnav\Delta\langle\rho\rangle\propto t^{n_{\mathrm{av}}} with nav=2+αn_{\mathrm{av}}=2+\alpha for t≪1t\ll1, while the percolation temperature is still the median, tp=2−1/(1+α)t_p=2^{-1/(1+\alpha)}, and ⟨ρ⟩(0)=(1+α)/(2+α)\langle\rho\rangle(0)=(1+\alpha)/(2+\alpha). Eliminating α\alpha,

nav=1+ln⁡2ln⁡(Tc,max⁡/Tp)(17)n_{\mathrm{av}}=1+\frac{\ln2}{\ln(T_{c,\max}/T_p)} \tag*{(17)}

The exponent approaches nav=1n_{\mathrm{av}}=1 only as the onset of screening is significantly suppressed: nav=1.5n_{\mathrm{av}}=1.5 for Tc,max⁡/Tp=4T_{c,\max}/T_p=4, 1.251.25 for 1616, and 1.11.1 for 10241024. The log-uniform distribution, P∝1/tiP\propto1/t_i on [tmin⁡,1][t_{\min},1], is the limit of the power-law distribution when α→−1\alpha\to-1. It gives Δ⟨ρ⟩∝t/ln⁡(1/tmin⁡)\Delta\langle\rho\rangle\propto t/\ln(1/t_{\min}) for tmin⁡≪tt_{\min}\ll t, but tp=tmin⁡t_p=\sqrt{t_{\min}}. For example, for tmin⁡=0.01t_{\min}=0.01, tp=0.1t_p=0.1: the film screens only below 0.1Tc,max⁡0.1T_{c,\max}, well below Tc,max⁡/3T_{c,\max}/3, the usual upper limit of the low-temperature range used to fit λ(T)\lambda(T).

P(ti)P(t_i)tpt_p⟨ρ⟩(0)\langle\rho\rangle(0)ρeff(0)\rho_{\mathrm{eff}}(0)navn_{\mathrm{av}}nn
uniform (tmin⁡=0t_{\min}=0)0.50.500.38521.66–1.73
∝ti−1/2\propto t_i^{-1/2}0.250.330.1721.51.32–1.47
∝ti−3/4\propto t_i^{-3/4}0.06250.200.0421.251.11–1.23
log-uniform0.0320.150.0311.151.24–1.36
d: uniform (tmin⁡=0t_{\min}=0)0.50.500.38511.19–1.34
d: ∝ti−1/2\propto t_i^{-1/2}0.250.330.17211.09–1.26

Table 2. Low-temperature penetration depth of mosaics with non-uniform TcT_c distributions (s-wave unless marked d). navn_{\mathrm{av}}: exponent of Δ⟨ρ⟩=⟨ρ⟩(0)−⟨ρ⟩(t)\Delta\langle\rho\rangle=\langle\rho\rangle(0)-\langle\rho\rangle(t), Eq. (17) (fit for the log-uniform case). nn: exponent of Δλeff∝tn\Delta\lambda_{\mathrm{eff}}\propto t^n from fits over the windows 0.05–0.3, 0.1–0.5 and 0.05–0.6 of tpt_p (range of the three values). Log-uniform is ∝1/ti\propto1/t_i on [10−3,1][10^{-3},1].

Of course, the average navn_{\mathrm{av}} is not what is measured. To see how the mosaic domains modify the behavior, we computed ρeff\rho_{\mathrm{eff}} for different P(ti)P(t_i) distributions with the same geometry realizations at 37 temperatures up to 1.1tp1.1t_p, as summarized in Table II. For the s-wave powerlaw distributions, as for the uniform one, the film exponent lies below navn_{\mathrm{av}}, because the weak domains that lose their condensate carry u>1u>1. In units of its own onset of screening, the s-wave mosaic with P∝ti−3/4P\propto t_i^{-3/4} gives Δλeff∝t1.1−1.2\Delta\lambda_{\mathrm{eff}}\propto t^{1.1-1.2}, as close to linear as the d-wave mosaic with a uniform distribution. However, in this case, the onset of screening lies 16 times below the highest local Tc,maxT_{\mathrm{c,max}}, half of the domains have their local TcT_{\mathrm{c}} between TpT_p and 16Tp16T_p, and ρeff(0)=0.04\rho_{\mathrm{eff}}(0)=0.04, i.e., λeff(0)=4.9λ1(0)\lambda_{\mathrm{eff}}(0)=4.9\lambda_1(0). The log-uniform case does not produce closer to linear behavior either, because its cutoff makes λeff\lambda_{\mathrm{eff}} activated again below t≈tmin⁡=0.03tpt\approx t_{\min}=0.03t_p.

Still, the nn values in the range of 1.1-1.3 may look close to linear, especially with the always present noise in the experimental data. Therefore, the exponent alone cannot exclude an s-wave mosaic. The decisive quantity is the ratio Tc,max/TpT_{\mathrm{c,max}}/T_p between the highest local TcT_{\mathrm{c}}, seen by local probes (not only magnetic, but including, for example, tunneling spectroscopy), and the onset of macroscopic screening, TpT_p. Since these quite specialized measurements may not be available for every film studied, transport measurements can be also used by detecting the onset of the resistive transition, which will be visible even for non-percolating clusters. Importantly, the offset of the resistive transition, R=0R=0, must be close to percolation threshold, tpt_p, and can be compared with the onset of macroscopic screening seen by the measurements of χ(T)\chi(T). For example, s-wave n≤1.5n\leq1.5 requires Tc,max/Tp≳4T_{\mathrm{c,max}}/T_p\gtrsim4, and n≈1.2n\approx1.2 requires Tc,max/Tp≈16T_{\mathrm{c,max}}/T_p\approx16, both easily resolvable.

Discussion

Behavior near the onset of the transition.

In the developed TcT_{\mathrm{c}}-mosaic model, the superconducting transition seen in whole-sample or part-sample mutual-inductance measurements of χ(T)\chi(T), is a percolation transition of the global superfluid density. Its onset is at the median of the local TcT_{\mathrm{c}} distribution, TpT_p, not at the maximum, Tc,maxT_{\mathrm{c,max}}. A uniform film has ρ∝(Tc−T)\rho\propto(T_{\mathrm{c}}-T) near TcT_{\mathrm{c}}. In weak screening, ∣χ∣≈(3π/32)(a/Λ)χ0∝ρ|\chi|\approx(3\pi/32)(a/\Lambda)\chi_0\propto\rho, so χ\chi starts linearly from TcT_{\mathrm{c}} and dχ/dTd\chi/dT jumps at TcT_{\mathrm{c}}, which is a kink in χ(T)\chi(T). The mosaic, on the other hand, has ρeff∝(tp−t)μ\rho_{\mathrm{eff}}\propto(t_p-t)^\mu in the critical region, with μ>1\mu>1 as for the conductivity of two-dimensional percolating networks (μ≈1.3\mu\approx1.3 asymptotically [16, 17]; effective exponents fitted over a wider range near tpt_p are larger), so dχ/dTd\chi/dT rises continuously from zero as shown in Figure 3(c,Figure 3d). The onset is a soft kink in the critical region and a rounded foot outside it. Below the onset, the curve is broader than that of the uniform film (tmin⁡=1t_{\min}=1 in Figure 3), because superperconductivity strengthens gradually as more domains join in and as each domain’s own ρ\rho grows. For tmin⁡=0t_{\min}=0 the 50% point of χ\chi is at t50=0.36t_{50}=0.36 (s-wave) and 0.33 (d-wave) for an onset at tp=0.5t_p=0.5. Regarding the steepness of χ(T)\chi(T) below the transition, because Λeff\Lambda_{\mathrm{eff}} crosses aa only when ρeff\rho_{\mathrm{eff}} drops to 0.033, χ(T)\chi(T) rises steeply below the onset and then saturates. The strong-screening plateau hides most of the low-temperature variation of ρeff\rho_{\mathrm{eff}}, which is visible only through the logarithm in 1−F1-F. Above the percolation transition, there is no measurable magnetic susceptibility (neglecting diamagnetism of non-percolating clusters which do not have large-loop currents). Yet, superconductivity starting at Tc,maxT_{\mathrm{c,max}} can still be clearly visible to local probes when they focus on the domains with maximum screening. The same transition can be detected as the onset in resistivity.

Therefore, analysis of the measured χ(T)\chi(T) and resistivity near the apparent transition, as well as comparison with local probes (if available) can be used to determine parameters of the TcT_{\mathrm{c}}-mosaic.

Distinguishing effect of inhomogeneity from genuine nodes.

For s-wave domains, inhomogeneity alone may produce a λ(T)\lambda(T) that does not saturate only if a finite fraction of the film has a local TcT_{\mathrm{c}} near zero, tmin⁡→0t_{\min}\to0. However, the same sample will show a greatly reduced onset of the transition, tp≤0.5t_p\leq0.5 for the uniform distribution and even lower for distributions weighted toward Tc→0T_{\mathrm{c}}\to0 (Sec. IV D), and a strongly suppressed ρeff(0)\rho_{\mathrm{eff}}(0).

Line nodes in uniformly scaled d-wave domains produce a linear term in λ(T)\lambda(T) whose relative slope is enhanced by ⟨u2⟩/ρeff(0)\langle u^2\rangle/\rho_{\mathrm{eff}}(0), Eq. (15). Local probes with resolution better than ℓ\ell measure single-domain ρ(T)\rho(T) and are insensitive to the network. Comparing the local and whole-sample λ(T)\lambda(T) on the same film can therefore separate the two contributions.

Assumptions and limitations of the TcT_c–mosaic model. The developed TcT_c–mosaic model has limitations that follow from the assumptions made: (1) The calculation is done in the mean-field, local London approximation. This requires ξ≪ℓ\xi\ll\ell and excludes nonlocal effects in clean nodal superconductors. (2) The domains are independent, with uncorrelated TcT_c. (3) There is no coupling through normal domains. Josephson or proximity coupling through them would connect clusters above tpt_p and increase the transition onset. For domains of size ℓ=20 μm\ell= 20\,\mu\mathrm{m}, much larger than the proximity length (≲0.1 μm\lesssim0.1\,\mu\mathrm{m}), this coupling is negligible. (4) Phase fluctuations are not included. Near tpt_p, the superfluid density of the mosaic network is small and vanishes continuously at tpt_p. Therefore, the Berezinskii–Kosterlitz–Thouless transition, at which the Pearl length reaches the universal value Λ(TBKT)=Φ024πμ0kBTBKT)\Lambda(T_{\mathrm{BKT}}) = \frac{\Phi_0^2}{4\pi\mu_0 k_{\mathrm{B}}T_{\mathrm{BKT}})} [32, 33, 34, 35, 36], always lies below tpt_p, and vortex–antivortex unbinding would further reduce the onset of screening.

Conclusions

Within the presented TcT_c–mosaic model, a superconducting film made entirely of domains with different critical temperatures screens a perpendicular field only below the percolation threshold temperature, TpT_p, of its superconducting domains, which is the median local TcT_c. Above TpT_p, only very weak residual macroscopic diamagnetic response of individual non-percolating clusters remains and is neglected in this model.

Fully gapped ss-wave domains with a positive lower cutoff, tmin⁡>0t_{\min} > 0, retain an asymptotically (at t→0t \to0) activated penetration depth, but its temperature interval shrinks with decreasing tmin⁡t_{\min}. Extending the TcT_c distribution to zero, tmin⁡→0t_{\min} \to0, removes this activated behavior while substantially suppressing the superfluid density and an apparent transition temperature, TpT_p. For uniform distributions of TcT_c, the fully gapped mosaic does not yield a TT-linear behavior over any temperature window: Δλeff∝tn\Delta\lambda_{\mathrm{eff}} \propto t^n with n=1.6n = 1.6–1.81.8 for tmin⁡=0t_{\min} = 0 and n>4n > 4 for tmin⁡=0.2t_{\min} = 0.2. For any general distribution, the area-averaged superfluid density approaches its zero-temperature value with zero slope, without a TT-linear term, and for power-law distributions the exponent of ⟨ρ⟩(0)−⟨ρ⟩(t)\langle\rho\rangle(0) - \langle\rho\rangle(t) is tied to the onset of screening, nav=1+ln⁡2/ln⁡(Tc,max⁡/Tp)n_{\mathrm{av}} = 1 + \ln2/\ln(T_{c,\max}/T_p). The behavior comes closer to linear than the area average, but only in units of TpT_p and only when TpT_p lies far below the highest local TcT_c: n≤1.5n \le1.5 requires Tc,max⁡/Tp≳4T_{c,\max}/T_p \gtrsim4, and n≈1.2n \approx1.2 requires Tc,max⁡/Tp≈16T_{c,\max}/T_p \approx16 and ρeff(0)≈0.04\rho_{\mathrm{eff}}(0) \approx0.04 (Table II).

Therefore, a TT-linear λ(T)\lambda(T) measured on a film whose onset of macroscopic screening is within a factor of a few of the highest local TcT_c, as seen by local probes or by the onset of the resistive transition, cannot be explained by TcT_c mosaicity of a fully gapped superconductor.

Nodal domains retain a linear low-temperature term (or whatever power law the individual domains have, set by impurity scattering) whose relative slope is enhanced by inhomogeneity.

These conclusions apply to the analysis of the magnetic susceptibility of the whole sample measured by global AC susceptibility and to the mutual-inductance technique, where the excitation field varies on millimeter scales, much larger than ℓ\ell, so the same homogenized Δλeff\Delta\lambda_{\mathrm{eff}} applies.

Acknowledgments

I thank Bicky Moirangthem for discussions. Claude AI was used to develop Python code for numerical calculations, which was then manually validated. This work was supported by the U.S. Department of Energy, Office of Science, National Quantum Information Science Research Centers, Superconducting Quantum Materials and Systems Center (SQMS), under Contract No. 89243024CSC000002. Ames National Laboratory is supported by the U.S. Department of Energy (DOE), Office of Science, Basic Energy Sciences (BES), Materials Science & Engineering Division (MSED) and is operated by Iowa State University for the U.S. DOE under contract DE-AC02-07CH11358.

References

  1. [1]R. Prozorov and R. W. Giannetta, Magnetic penetration depth in unconventional superconductors, Supercond. Sci. Technol. 19, R41 (2006).DOI
  2. [2]R. Prozorov and V. G. Kogan, London penetration depth in iron-based superconductors, Rep. Prog. Phys. 74, 124505 (2011).DOI
  3. [3]C. V. Topping and S. J. Blundell, AC susceptibility as a probe of low-frequency magnetic dynamics, J. Phys.: Condens. Matter 31, 013001 (2019).DOI
  4. [4]R. Giannetta, A. Carrington, and R. Prozorov, London penetration depth measurements using tunnel diode resonators, Journal of Low Temperature Physics 208, 119 (2022).DOI
  5. [5]B. S. Moirangthem, K. R. Joshi, Z.-J. Yan, P. Xiao, L.-K. Lai, M. A. Tanatar, C.-Z. Chang, and R. Prozorov, Anisotropic interface-confined superconductivity in FeTe-based heterostructures, arXiv:2609.38148 10.48550/arXiv.2609.38148 (2026).arxiv.org/abs/2609.38148
  6. [6]A. T. Fiori, A. F. Hebard, P. M. Mankiewich, and R. E. Howard, Penetration depths of high-Tc films measured by two-coil mutual inductances, Appl. Phys. Lett. 52, 2165 (1988).
  7. [7]S. J. Turneaure, A. A. Pesetski, and T. R. Lemberger, Numerical modeling and experimental considerations for a two-coil apparatus to measure the complex conductivity of superconducting films, Journal of Applied Physics 83, 4334 (1998).
  8. [8]Z.-J. Yan, H. Rong, Y. Luo, Y. Zhao, P. Xiao, Z. Wang, L.-K. Lai, A. G. Wang, Z. Xi, Y. Li, X. Wei, K. Wang, B. Yan, C.-K. Shih, and C.-Z. Chang, Interface-Confined Superconductivity with Thickness-Independent Superfluid Stiffness in (Pb,Sn)Te/FeTe Bilayers (2026), 2607.17539 [cond-mat.supr-con].
  9. [9]R. Prozorov, Magnetic susceptibility of superconducting films in the meissner-london regime (2026), 2610.20455 [cond-mat.supr-con].arxiv.org/abs/2610.20455
  10. [10]C. Li, Z.-J. Yan, Y. Ge, Z. Wang, B. Xia, S. Paolini, P. Xiao, L.-K. Lai, J. Song, A. R. Kaczmarek, L. Min, K. Yasuda, P. J. Hirschfeld, J. Yu, C.-Z. Chang, and K. C. Nowack, Microscopic origin of a two-dimensional superconductor in stoichiometric FeTe, arXiv:2609.08116 10.48550/arXiv.2609.08116 (2026).
  11. [11]I. Iguchi, T. Yamaguchi, and A. Sugimoto, Diamagnetic activity above Tc as a precursor to superconductivity in La2-xSrxCuO4 thin films, Nature 412, 420 (2001).pubmed.ncbi.nlm.nih.gov/11473312
  12. [12]G. M. Ferguson, H. P. Nair, N. J. Schreiber, L. Miao, K. M. Shen, D. G. Schlom, and K. C. Nowack, Local magnetic response of superconducting Sr 2 RuO 4 thin films, Physical Review B 110, 144510 (2024).
  13. [13]H. T. Yi, X. Yao, D. Jain, Y.-T. Chan, A.-H. Chen, M. Bra hlek, K. Kisslinger, K. Du, M.-G. Han, Y. Zhu, W. Wu, S.-W. Cheong, and S. Oh, Universal Superconductivity in FeTe and All-Iron-Based Ferromagnetic Superconductor Heterostructures, Advanced Functional Materials 35, 2418259 (2025).DOI
  14. [14]Z. Li, R. Zhang, M. Xu, K. Liang, Y. Zhao, Q. He, Q. Zhou, B. Chen, P. Zhang, K. Yao, H. Yao, L. Qiao, and Y. Wang, Correlation between superfluid density and transition temperature in infinite-layer nickelate superconductor Nd1−x Srx NiO2, National Science Review 13, nwag068 (2026).DOI
  15. [15]S. Kirkpatrick, Percolation and conduction, Rev. Mod. Phys. 45, 574 (1973).DOI
  16. [16]A. L. Efros and B. I. Shklovskii, Critical behaviour of conductivity and dielectric constant near the metal-non-metal transition threshold, Phys. Status Solidi B 76, 475 (1976).DOI
  17. [17]D. Stauffer and A. Aharony, Introduction to Percolation Theory, 2nd ed. (Taylor & Francis, London, 1994).
  18. [18]A. Okabe, B. Boots, K. Sugihara, and S. N. Chiu, Spatial Tessellations: Concepts and Applications of Voronoi Diagrams, 2nd ed. (Wiley, Chichester, 2000).
  19. [19]C. C. Homes et al, A universal scaling relation in high-temperature superconductors, Nature 430, 539 (2004).
  20. [20]V. G. Kogan, Homes scaling and BCS, Phys. Rev. B 87, 220507 (2013).DOI
  21. [21]M. Tinkham, Introduction to Superconductivity, 2nd ed. (McGraw-Hill, New York, 1996).
  22. [22]P. J. Hirschfeld and N. Goldenfeld, Effect of strong scattering on the low-temperature penetration depth of a d-wave superconductor, Phys. Rev. B 48, 4219 (1993).pubmed.ncbi.nlm.nih.gov/10008890
  23. [23]V. G. Kogan, R. Prozorov, and V. Mishra, London penetration depth and pair breaking, Phys. Rev. B 88, 224508 (2013).DOI
  24. [24]J. Pearl, Current distribution in superconducting films carrying quantized fluxoids, Appl. Phys. Lett. 5, 65 (1964).DOI
  25. [25]J. B. Keller, A theorem on the conductivity of a composite medium, J. Math. Phys. 5, 548 (1964).DOI
  26. [26]A. M. Dykhne, Conductivity of a two-dimensional two-phase system, Sov. Phys. JETP 32, 63 (1971), [Zh. Eksp. Teor. Fiz. 59, 110 (1970)].
  27. [27]J. R. Clem and A. Sanchez, Hysteretic ac losses and susceptibility of thin superconducting disks, Phys. Rev. B 50, 9355 (1994).pubmed.ncbi.nlm.nih.gov/9974983
  28. [28]Y. Saad, Iterative Methods for Sparse Linear Systems, 2nd ed. (SIAM, Philadelphia, 2003).DOI
  29. [29]X. S. Li, An overview of SuperLU: Algorithms, implementation, and user interface, ACM Trans. Math. Softw. 31, 302 (2005).
  30. [30]T. A. Davis, J. R. Gilbert, S. I. Larimore, and E. G. Ng, A column approximate minimum degree ordering algorithm, ACM Trans. Math. Softw. 30, 353 (2004).
  31. [31]B. Bollobás and O. Riordan, The critical probability for random Voronoi percolation in the plane is 1/2, Probab. Theory Relat. Fields 136, 417 (2006).DOI
  32. [32]V. L. Berezinskii, Destruction of long-range order in one-dimensional and two-dimensional systems possessing a continuous symmetry group. II, Quantum systems, Zh. Eksp. Teor. Fiz. 61, 1144 (1971), [Sov. Phys. JETP 34, 610 (1972)].
  33. [33]J. M. Kosterlitz and D. J. Thouless, Ordering, metastability and phase transitions in two-dimensional systems, J. Phys. C 6, 1181 (1973).
  34. [34]D. R. Nelson and J. M. Kosterlitz, Universal jump in the superfluid density of two-dimensional superfluids, Phys. Rev. Lett. 39, 1201 (1977).DOI
  35. [35]M. R. Beasley, J. E. Mooij, and T. P. Orlando, Possibility of vortex-antivortex pair dissociation in two-dimensional superconductors, Phys. Rev. Lett. 42, 1165 (1979).DOI
  36. [36]A. F. Hebard and A. T. Fiory, Evidence for the Kosterlitz-Thouless transition in thin superconducting aluminum films, Phys. Rev. Lett. 44, 291 (1980).

Paper details

vlnc:W_rwh39gyz
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