The perfectly linear temperature dependence of the electrical resistivity observed as T → 0 in a variety of metals close to a quantum critical point1,2,3,4 is a major puzzle of condensed-matter physics5. Here we show that T-linear resistivity as T → 0 is a generic property of cuprates, associated with a universal scattering rate. We measured the low-temperature resistivity of the bilayer cuprate Bi2Sr2CaCu2O8+δ and found that it exhibits a T-linear dependence with the same slope as in the single-layer cuprates Bi2Sr2CuO6+δ (ref. 6), La1.6−xNd0.4SrxCuO4 (ref. 7) and La2−xSrxCuO4 (ref. 8), despite their very different Fermi surfaces and structural, superconducting and magnetic properties. We then show that the T-linear coefficient (per CuO2 plane), A1□, is given by the universal relation A1□TF = h/2e2, where e is the electron charge, h is the Planck constant and TF is the Fermi temperature. This relation, obtained by assuming that the scattering rate 1/τ of charge carriers reaches the Planckian limit9,10, whereby ħ/τ = kBT, works not only for hole-doped cuprates6,7,8,11,12 but also for electron-doped cuprates13,14, despite the different nature of their quantum critical point and strength of their electron correlations.
In conventional metals, the electrical resistivity ρ(T) normally varies as T2 in the limit T → 0, where electron–electron scattering dominates, in accordance with Fermi-liquid theory. However, close to a quantum critical point (QCP) where a phase of antiferromagnetic order ends, ρ(T) ~ Tn, with n < 2.0. Most striking is the observation of a perfectly linear T dependence ρ(T) = ρ0 + A1T as T → 0 in several very different materials, when tuned to their magnetic QCP; for example, the quasi-one-dimensional (1D) organic conductor (TMTSF)2PF6 (ref. 4), the quasi-2D ruthenate Sr3Ru2O7 (ref. 3) and the 3D heavy-fermion metal CeCu6 (ref. 1). This T-linear resistivity as T → 0 has emerged as one of the major puzzles in the physics of metals5, and while several theoretical scenarios have been proposed15, no compelling explanation has been found.
In cuprates, a perfect T-linear resistivity as T → 0 has been observed (once superconductivity is suppressed by a magnetic field) in two closely related electron-doped materials, Pr2−xCexCuO4±δ (PCCO)2,16,17 and La2−xCexCuO4 (LCCO)13,14, and in three hole-doped materials: Bi2Sr2CuO6+δ (ref. 6), La2−xSrxCuO4 (LSCO)8 and La1.6−xNd0.4SrxCuO4 (Nd-LSCO)7,11,12. On the electron-doped side, T-linear resistivity is seen just above the QCP16 where antiferromagnetic order ends18 as a function of x, and as such it may not come as a surprise. On the hole-doped side, however, the doping values where ρ(T) = ρ0 + A1T as T → 0 are very far from the QCP where long-range antiferromagnetic order ends (pN ~ 0.02); for example, at p = 0.24 in Nd-LSCO (Fig. 1a) and in the range p = 0.21–0.26 in LSCO (Fig. 1b). Instead, these values are close to the critical doping where the pseudogap phase ends (that is, at p* = 0.23 ± 0.01 in Nd-LSCO (ref. 11) and at p* ~ 0.18–0.19 in LSCO (ref. 8)), where the role of antiferromagnetic spin fluctuations is not clear. In Bi2201, p* is farther still (see Supplementary Section 10).
To make progress, several questions must be answered. Is T-linear resistivity as T → 0 in hole-doped cuprates limited to single-layer materials with low Tc, or is it generic? Why is ρ(T) = ρ0 + A1T as T → 0 seen in LSCO over an anomalously wide doping range8? Is there a common mechanism linking cuprates to the other metals where ρ ~ T as T → 0?
To establish the universal character of T-linear resistivity in cuprates, we have turned to Bi2Sr2CaCu2O8+δ (Bi2212). While Nd-LSCO and LSCO have essentially the same single electron-like diamond-shaped Fermi surface at p > p* (refs 19,20), Bi2212 has a very different Fermi surface, consisting of two sheets, one of which is also diamond-like at p > 0.22, but the other is much more circular21 (see Supplementary Section 1). Moreover, the structural, magnetic and superconducting properties of Bi2212 are very different to those of Nd-LSCO and LSCO: a stronger 2D character, a larger gap to spin excitations, no spin-density-wave order above p ~ 0.1 and a much higher superconducting Tc.
We measured the resistivity of Bi2212 at p = 0.23 by suppressing superconductivity with a magnetic field of up to 58 T. At p = 0.23, the system is just above its pseudogap critical point (p* = 0.22 (ref. 22); see Supplementary Section 2). Our data are shown in Fig. 2. The raw data at H = 55 T reveal a perfectly linear T dependence of ρ(T) down to the lowest accessible temperature (Fig. 1a). Correcting for the magnetoresistance (see Methods and Supplementary Section 3), as was done for LSCO (ref. 8), we find that the T-linear dependence of ρ(T) seen in Bi2212 at H = 0 from T ~ 120 K down to Tc simply continues to low temperature, with the same slope A1 = 0.62 ± 0.06 μΩ cm K−1 (Fig. 2b). Measured per CuO2 plane, this gives A1□ ≡ A1/d = 8.0 ± 0.9 Ω K−1, where d is the (average) separation between CuO2 planes. Remarkably, this is the same value, within error bars, as measured in Nd-LSCO at p = 0.24, where A1□ = 7.4 ± 0.8 Ω K−1 (see Table 1).
The observation of T-linear resistivity in those two cuprates shows that it is robust against changes in the shape, topology and multiplicity of the Fermi surface. By contrast, the Hall coefficient RH is not. In Fig. 2d, we compare RH(T) in Bi2212 and in Nd-LSCO (and PCCO). We see strong differences, brought about by the different anisotropies in either the inelastic scattering or the Fermi surface, or both23. Nevertheless, ρ(T) is perfectly linear in both cases. Moreover, the coefficient A1□ is the same despite the very different spectra of low-energy spin fluctuations, gapped in Bi2212 (ref. 24) and ungapped in Nd-LSCO (ref. 25). We conclude that a T-linear resistivity as T → 0 is a generic and robust property of cuprates.
Note that ρ(T) deviates from pure T-linearity above a certain temperature, and in this high-T regime a generic evolution has also been found in LSCO (ref. 26), with ρ(T) ~ A1T + A2T2. Here we focus strictly on the low-T regime of pure T-linear resistivity (see Supplementary Section 12). In this regime, and close to the QCP of BaFe2(As1−xPx)2 (at x = 0.31), an empirical scaling relationship between applied magnetic field and temperature has been proposed27, but this scaling does not work very well in Bi2212 (see Supplementary Section 11).
We now investigate the strength of the T-linear resistivity; that is, the magnitude of A1. In Fig. 3b, we plot A1□ versus p for hole-doped cuprates. We see from the LSCO data8 that A1□ increases with decreasing p (Fig. 1b), from A1□ ~ 8 Ω K−1 at p = 0.26 to A1□ ~ 15 Ω K−1 at p = 0.21 (see Supplementary Table 2 in Supplementary Section 13 and Methods). In Nd-LSCO, we see a similar increase (Figs. 1c and 3b), when pressure12 is used to suppress the onset of the pseudogap at p = 0.22 and p = 0.23 (see Supplementary Section 4). In Fig. 1d, we present our data on PCCO at x = 0.17 (see also Supplementary Section 5), and compare with previous data on LCCO (ref. 14; Supplementary Section 6). In Fig. 4b, we plot A1□ versus x for electron-doped cuprates, and see that A1□ also increases with decreasing x, from A1□ ~ 1.5 Ω K−1 at x = 0.17 to A1□ ~ 3 Ω K−1 at x = 0.15 (see Supplementary Table 4 in Supplementary Section 13 and Methods). Note that these values are five times smaller than in hole-doped cuprates.
To summarize: A1□ increases as the doping is reduced in both hole-doped and electron-doped cuprates; A1□ is much larger in hole-doped cuprates; T-linear resistivity as T → 0 is observed over a range of doping, not just at one doping; T-linear resistivity does not depend on the nature of the inelastic scattering process (hole-doped versus electron-doped) or on the topology of the Fermi surface (LSCO versus NCCO, Bi2212 versus Nd-LSCO; Supplementary Section 1).
To explain these experimental facts, we consider the empirical observation that the strength of the T-linear resistivity for several metals is approximately given by a scattering rate that has a universal value, namely ħ/τ = kBT (ref. 10), and test it in cuprates. This observation suggests that a T-linear regime will be observed whenever 1/τ reaches its Planckian limit, kBT/ħ, irrespective of the underlying mechanism for inelastic scattering9. In the following, we use a standard Fermi-liquid approach to extract effective masses and inelastic scattering rates, as in ref. 10. In the simple case of an isotropic Fermi surface, the connection between ρ and τ is given by the Drude formula, ρ = (m*/ne2) (1/τ), where n is the carrier density and m* is the effective mass. Thus, when ρ(T) = ρ0 + A1T, then A1 = (m*/n e2)(1/τ)(1/T) = α(m*/n)(kB/e2ħ), with ħ/τ ≡ αkBT. In two dimensions, this can be written succinctly as:
where TF = (πħ2/kB)(nd/m*) is the Fermi temperature.
Let us first evaluate α in electron-doped cuprates, where the Drude formula is expected to work well, since their single Fermi surface is highly 2D and circular (in the overdoped region28; see Supplementary Section 1). Quantum oscillations in Nd2−xCexCuO4 (NCCO) provide a direct and precise measurement of n and m* in electron-doped cuprates29,30. The Luttinger rule sets the carrier density to be n = (1 − x)/(a2d), given precisely by the oscillation frequency F = nd(h/2e), where x is the number of doped electrons per Cu atom and a is the in-plane lattice constant. In Fig. 4a, we see that m* increases from 2.3 m0 at x = 0.173 to 3.0 m0 at x = 0.151, where m0 is the bare electron mass (see Supplementary Table 3 in Supplementary Section 13 and Methods). This increasing value is consistent, within error bars, with specific heat data in PCCO at x = 0.15, where γ = 5.5 ± 0.4 mJ K−2 mol−1 (ref. 31), which yields m* = 3.6 ± 0.3 m0 (see equation (2) below). We use n and m* to estimate TF and then plot, in Fig. 4b, the value of A1□ predicted by equation (1), for α = 1 (solid line in Fig. 4b; Supplementary Table 3 in Supplementary Section 13). Comparison with the measured values of A1□ in PCCO (red hexagon in Fig. 4b) and in LCCO (blue circles in Fig. 4b), listed in Supplementary Table 4 (see Supplementary Section 13 and Methods), shows that the scattering rate in electron-doped cuprates is given by ħ/τ = αkBT, with α = 1.0 ± 0.3; that is, the Planckian limit is observed, within experimental error bars.
Let us now turn to hole-doped cuprates. Here our quantitative estimates will be more approximate, since Fermi surfaces are not circular but diamond-shaped (Supplementary Section 1), but we are looking for a large effect (factor ~5 in A1□ relative to electron-doped materials) and a qualitative trend (increase in A1□ as p is reduced towards p*). In the absence of quantum oscillation data for Bi2212, LSCO, Nd-LSCO and Bi2201, we estimate m* from specific heat data, since in two dimensions the specific heat coefficient γ is directly related to m*:
for a single Fermi surface, where NA is Avogadro’s number. This connection between m* and γ was nicely confirmed by quantum oscillations in Tl2Ba2CuO6+δ at p ~ 0.3, where m* = 5.2 ± 0.4 m0 and γ = 7 ± 1 mJ K−2 mol−1 (ref. 32). In Bi2212, γ = 12 ± 2 mJ K−2 per mol-Cu at p = 0.22 = p* (ref. 33; see Supplementary Section 8), giving m* = 8.4 ± 1.6 m0 (equation (2)). Applying equation (1), with n(a2d) = 1 − p = 0.77 (for an electron-like Fermi surface; Supplementary Section 1), the Planckian limit predicts A1□ = 7.4 ± 1.4 Ω K−1, while we measured A1□ = 8.0 ± 0.9 Ω K−1, so that α = 1.1 ± 0.3 (Table 1).
In LSCO, γ increases from γ = 6.9 ± 1 mJ K−2 mol−1 at p = 0.33 (ref. 34) to γ = 14 ± 2 mJ K−2 mol−1 at p = 0.26 (ref. 35), showing that m* increases with reduced doping also in hole-doped cuprates (solid line in Fig. 3a). Applying equation (1) to LSCO data at p = 0.26, using n(a2d) = 1 − p = 0.74 and m* = 9.8 ± 1.7 m0 (equation (2); Supplementary Table 1 in Supplementary Section 13), the Planckian limit predicts A1□ = 8.9 ± 1.8 Ω K−1, while we see A1□ = 8.2 ± 1.0 Ω K−1 (Fig. 1b and Supplementary Table 2 in Supplementary Section 13), so that α = 0.9 ± 0.3 (Table 1).
In Nd-LSCO, an increase in m* has also been observed in recent specific heat measurements36, from γ = 5.4 ± 1 mJ K−2 mol−1 at p = 0.40 to γ = 11 ± 1 mJ K−2 mol−1 at p = 0.27 (Fig. 3a). At p = 0.24, the electronic specific heat Cel varies as Cel/T ~ log(1/T), which complicates the estimation of m*. Taking the mean value between Cel/T = 12 mJ K−2 mol−1 at 10 K and Cel/T = 22 mJ K−2 mol−1 at 0.5 K (ref. 36), we get m* = 12 ± 4 m0 and hence α = 0.7 ± 0.4, consistent with the Planckian limit for a third hole-doped material. See Table 1 for a summary of the numbers.
Finally, a stringent test of whether the Planckian limit operates in cuprates is provided by Bi2201, since in this particular cuprate the pseudogap critical point that controls T-linear scattering occurs at a much higher doping than in other cuprates, namely p* ~ 0.4 (see Supplementary Section 10). Despite this doubling of p* and the very different volume of the Fermi surface relative to Bi2212, LSCO and Nd-LSCO, we find that α = 1.0 ± 0.4 in Bi2201 (Table 1 and Supplementary Section 10).
In summary, our estimations reveal that the scattering rate responsible for the T-linear resistivity in PCCO, LCCO, Bi2212, LSCO, Nd-LSCO and Bi2201 tends to the same universal value, namely ħ/τ = αkBT, with α = 1.0 (Table 1). A constant value of α in equation (1) implies that A1□ ~ 1/TF, so that, in essence, A1□ ~ m*. This explains why the slope of the T-linear resistivity is much larger in hole-doped than in electron-doped cuprates, since the effective mass is much higher in the former (Fig. 3a versus Fig. 4a). It also explains why A1□ increases in LSCO when going from p = 0.26 to p = 0.21 (Fig. 1b) and in Nd-LSCO (under pressure) when going from p = 0.24 to p = 0.22 (Fig. 1c). Indeed, as shown in Fig. 3, A1□ (Fig. 3b) and m* (Fig. 3a) in LSCO and Nd-LSCO are seen to rise in tandem with decreasing p (we make the natural assumption that m* continues to rise until p reaches p* and that the pressure does not change the specific heat significantly above p* in Nd-LSCO). Moreover, a Planckian limit on scattering provides an explanation for the ‘anomalous’ range in doping over which ρ ~ A1T is observed in LSCO (ref. 8). As doping decreases below p ~ 0.33, scattering increases steadily until p* ~ 0.18–19, but the inelastic scattering rate 1/τ cannot exceed the Planckian limit, reached at p ~ 0.26. Thus, between p ~ 0.26 and p = p*, ρ(T) is linear and 1/τ saturates. The continuous increase of A1 below p ~ 0.26 can be understood if we assume that m* continues to increase in the range p* < p < 0.26 (ref. 36), since A1 ~ m*(1/τ) ~ m*. If one could lower p*, the range of T-linear resistivity would expand further. This is indeed what happens in Nd-LSCO when p* is lowered by applying pressure12 (Fig. 3b).
The fact that α ~ 1.0 in cuprates has far-reaching implications since other metals with T-linear resistivity as T → 0 also appear to have α ~ 1.0 (ref. 10). The case is particularly clear in the organic conductor (TMTSF)2PF6, a well-characterized single-band metal whose resistivity is perfectly T-linear as T → 0 (ref. 4), where α = 1.0 ± 0.3 (Table 1 and Supplementary Section 9). For such dramatically different metals as the quasi-1D organics and the cuprates—not to mention the heavy-fermion metals and the pnictides10—to all have quantitatively the same scattering rate in their respective T-linear regimes, there must be a fundamental and universal principle at play. Our findings support the idea9,10 that T-linear resistivity is achieved when the scattering rate hits the Planckian limit, given by ħ/τ = kBT, whatever the scattering process, whether by antiferromagnetic spin fluctuations or not. If Planckian dissipation is the fundamental principle, new theoretical approaches are needed to understand how it works37,38,39.
Our thin film of Bi2Sr2CaCu2O8+δ (Bi2212) was grown epitaxially at 740 °C on a SrTiO3 substrate by radiofrequency magnetron sputtering with O2/Ar gas and fully oxygen overdoped after deposition40. The film thickness was measured by deposition rate calibration, giving t = 240 ± 15 nm. The film was patterned by mechanical scribing (avoiding the need for a lithography resist) into the shape of a Hall bar consisting of two large pads (for current) connected by a narrow bridge (275 µm wide) between 2 couples of voltage pads distant by 1.15 mm for longitudinal and transverse resistance measurements. Six gold contacts were deposited by sputtering on the different pads and gold wires were attached with silver paint.
The superconducting transition temperature Tc = 50 K was determined as the temperature below which the zero-field resistance R = 0. The hole doping p is obtained from Tc, using the usual convention21,22, according to which our overdoped sample has a nominal doping p = 0.23. This means that its doping is just slightly above the end of the pseudogap phase22 (see Supplementary Section 2). It is also just above the Lifshitz transition where its anti-bonding band crosses the Fermi level to produce an electron-like diamond-shaped Fermi surface21 (see Supplementary Section 1).
Our thin films of Pr2−xCexCuO4±δ (PCCO) were grown by pulsed laser deposition on LSAT substrates under 200 mTorr of N2O using targets including an excess of Cu to suppress the growth of parasitic phases41. Films were then annealed for 4 min in vacuum. The film thickness was measured via the width of X-ray diffraction peaks, giving t = 230 ± 30 nm. A very small amount of parasitic phase was detected in the X-ray diffraction spectra. However, its impact on the cross-section of the films should be much smaller than the uncertainty coming from the thickness measurement. Six indium–silver contacts were applied in the standard geometry.
The superconducting transition temperature Tc = 13 K was determined as the temperature below which the zero-field resistance R = 0. The electron concentration is taken to be the cerium content, x = 0.17, with an error bar ±0.005. This means that our samples have a concentration slightly above the quantum critical point where the Fermi surface of PCCO is known to undergo a reconstruction by antiferromagnetic ordering16. The Fermi surface of NCCO at that doping could not be simpler: it is a single circular cylinder28 (see Supplementary Section 1).
Measurement of the longitudinal and transverse resistances
The longitudinal resistance Rxx and transverse (Hall) resistance Rxy of our Bi2212 film were measured in Toulouse in pulsed fields up to 58 T. The measurements were performed using a conventional 6-point configuration with a current excitation of 0.5 mA at a frequency of ~10 kHz. A high-speed acquisition system was used to digitize the reference signal (current) and the voltage drop across the sample at a frequency of 500 kHz. The data were post-analysed with software to perform the phase comparison. Data for the rise and fall of the field pulse were in good agreement, thus excluding any heating due to eddy currents. Tests at different frequencies showed excellent reproducibility.
Rxx and Rxy of our Bi2212 film were also measured in Orsay, at H = 0 and H = 9 T, respectively.
The longitudinal resistance Rxx values of our three PCCO films were measured in Sherbrooke in a zero field and in a steady field of 16 T.
Values of m* and A 1
The values of p and m* used in Fig. 3a are listed in Supplementary Table 1 (see Supplementary Section 13). For Nd-LSCO, the value of p with its error bar is taken from ref. 11. For LSCO, the value of p is taken from refs 34,35, and we assume the same error bar as for Nd-LSCO. The value of m* is obtained from the measured specific heat γ, via equation (2). For Nd-LSCO, the value of γ with its error bar is taken from ref. 36, except for p = 0.24, where we take the average between the electronic specific heat Ce/T at T = 10 K (12 mJ K−2 mol−1) and at T = 0.5 K (22 mJ K−2 mol−1), given that Ce/T is not constant at low T (ref. 36). For Bi2212 and LSCO, we estimate γ and its error bar from the data published in ref. 33 and in refs 34,35, respectively (see Supplementary Section 8). With these values of m*, we calculate TF = (πħ2/kB)(nd/m*), using n = (1 − p)/(a2d) since the Fermi surface of Bi2212, LSCO and Nd-LSCO is electron-like at the dopings considered here (see Supplementary Section 1). We then obtain the Planckian limit on the resistivity slope, namely A1□ = h/(2e2TF), whose values are listed in the last column of Supplementary Table 2 (see Supplementary Section 13) and plotted in Fig. 3b (solid black line). For Bi2201, the values of n, m* and A1 are given in Supplementary Section 10, with associated error bars and references.
The values of p and A1 used in Fig. 3b are listed in Supplementary Table 2 (see Supplementary Section 13). For Nd-LSCO, the value of p with its error bar is taken from ref. 11. For LSCO, the value of p is taken from refs 8,42, and we assume the same error bar as for Nd-LSCO. For Nd-LSCO, the value of A1 is obtained from a linear fit to the raw data in Fig. 1a (p = 0.24, at H = 16 T) and in Fig. 1c (p = 0.22 and 0.23, at H = 33 T and P = 2 GPa). Note that the magnetoresistance is very weak in Nd-LSCO. For example, at p = 0.24, A1 = 0.47 μΩ cm K−1 at H = 33 T (Fig. 1c) versus A1 = 0.49 μΩ cm K−1 at H = 16 T (Fig. 1a). For LSCO, the value of A1 is obtained from a linear fit to the raw data in Fig. 1b (p = 0.26, at H = 18 T) and to the magnetoresistance-corrected data in Supplementary Section 7 (p = 0.21 and 0.23). Note that the magnetoresistance in LSCO does not significantly change the slope A1 (Fig. 1b versus Supplementary Fig. 7). For Bi2212, the value of A1 is obtained from a linear fit to magnetoresistance-corrected data (Fig. 2b and Supplementary Section 3). The error bar on A1 is in all cases taken to be ±10%, the estimated uncertainty in measuring the geometric factor of small samples. The values of A1 listed in Supplementary Table 2 (see Supplementary Section 13) are used to obtain the experimental values of A1□ = A1/d that are plotted in Fig. 3b.
The values of x and m* used in Fig. 4a are listed in Supplementary Table 3 (see Supplementary Section 13). For NCCO, the value of x is obtained from the frequency F of quantum oscillations, measured precisely29,30 (see Supplementary Section 1), via x = 1 − (2eFa2/h). The value of m* is obtained directly from the high-frequency quantum oscillations, as reported (with error bar) in refs 29,30. At the lowest doping (for example, x = 0.15), the Fermi surface of NCCO is reconstructed by an antiferromagnetic order leading to small electron and hole pockets. The small gap between the pockets decreases as the doping increases and, at x = 0.17, angle-resolved photoemission spectroscopy data show that the large hole-like Fermi surface is recovered28. At doping levels x = 0.151 to 0.163, we assume that the effective mass of the high quantum oscillation frequency due to magnetic breakdown is representative of the large hole-like Fermi surface, as expected in semiclassical theory43. The continuous trend of decreasing effective mass in the doping range x = 0.151 to 0.163 is confirmed by the smaller (and smoothly connected) value at x = 0.173.
For PCCO, x is taken to be the cerium content, with an error bar ±0.005. Here m* is obtained from the measured specific heat γ, via equation (2), and the value of γ (with its error bar) is taken from ref. 31. With these values of m*, we calculate TF = (πħ2/kB)(nd/m*), using n = (1 − x)/(a2d) since the Fermi surface of NCCO and PCCO is hole-like (see Supplementary Section 1). The good agreement between the specific heat data in PCCO at x = 0.15 and the values of the effective mass in NCCO (Fig. 4a) further confirms the validity of our approach regarding the extraction of m*.
We then obtain the Planckian limit on the resistivity slope, namely A1□ = h/(2e2TF), whose values are listed in the last column of Supplementary Table 3 (see Supplementary Section 13) and plotted in Fig. 4b (solid black line).
The values of x and A1 used in Fig. 4b are listed in Supplementary Table 4 (see Supplementary Section 13). In all cases, x is taken to be the cerium content, with an error bar ±0.005. For PCCO at x = 0.17, the value of A1 is obtained from a linear fit to the raw data in Supplementary Section 5. Within error bars, the same value is measured in all three PCCO films, whether at H = 0 or at H = 16 T. For LCCO, the value of A1 is obtained from a linear fit to the raw data in Fig. 1d. Also listed in Supplementary Table 4 (see Supplementary Section 13) are the values of A1 obtained from a linear fit to the raw zero-field data in LCCO (see Supplementary Section 6). The error bar on A1 is ±15% for our PCCO film, the uncertainty in measuring the film thickness. We apply the same error bar for LCCO. The values of A1 listed in Supplementary Table 4 (see Supplementary Section 13), both in zero field and in finite field, are used to obtain the experimental values of A1□ = A1/d that are plotted in Fig. 4b (as filled and open circles, respectively).
The data that support the plots within this paper and other findings of this study are available from the corresponding authors (L.T. or C.P.) upon reasonable request.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors would like to thank K. Behnia, C. Bourbonnais, R. Greene, S. Hartnoll, N. Hussey, M.-H. Julien, D. Maslov, J. Paglione, S. Sachdev, A.-M. Tremblay and J. Zaanen for fruitful discussions. A portion of this work was performed at the LNCMI, a member of the European Magnetic Field Laboratory (EMFL). C.P. acknowledges funding from the French ANR SUPERFIELD, and the LABEX NEXT. P.F. and L.T. acknowledge support from the Canadian Institute for Advanced Research (CIFAR) and funding from the Natural Sciences and Engineering Research Council of Canada (NSERC), the Fonds de recherche du Québec - Nature et Technologies (FRQNT) and the Canada Foundation for Innovation (CFI). L.T. acknowledges support from a Canada Research Chair. This research was undertaken thanks in part to funding from the Canada First Research Excellence Fund. Part of this work was funded by the Gordon and Betty Moore Foundation’s EPiQS Initiative (grant GBMF5306 to L.T.).