Discovery of superconductivity in quasicrystal

Superconductivity is ubiquitous as evidenced by the observation in many crystals including carrier-doped oxides and diamond. Amorphous solids are no exception. However, it remains to be discovered in quasicrystals, in which atoms are ordered over long distances but not in a periodically repeating arrangement. Here we report electrical resistivity, magnetization, and specific-heat measurements of Al–Zn–Mg quasicrystal, presenting convincing evidence for the emergence of bulk superconductivity at a very low transition temperature of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{\rm c} \cong 0.05$$\end{document}Tc≅0.05 K. We also find superconductivity in its approximant crystals, structures that are periodic, but that are very similar to quasicrystals. These observations demonstrate that the effective interaction between electrons remains attractive under variation of the atomic arrangement from periodic to quasiperiodic one. The discovery of the superconducting quasicrystal, in which the fractal geometry interplays with superconductivity, opens the door to a new type of superconductivity, fractal superconductivity.

I n classical crystallography, a crystal was defined as a periodic arrangement of atoms with translational periodicity, leading to an infinitely extended crystal structure by aligning building blocks called unit cells; as an example, we illustrate a cubic unit cell (Fig. 1a), in which the corner and body-centered positions are occupied by the icosahedron. This traditional definition was forced to modify by the discovery of quasicrystal (QC) by Shechtman et al. 1 , which led to a paradigm shift in science. Nowadays, QC is understood as a structure that is long-range ordered (as manifested in the occurrence of sharp diffraction spots) but not periodic (Fig. 1b) [2][3][4] . Another characteristic of QC is the presence of a non-crystallographic rotational symmetry 2,3 ; whereas periodic crystals can possess only two-, three-, four-, and sixfold rotational symmetries, icosahedral QCs have fivefold symmetry (Fig. 1b). In recent years, cold atom gaseous QCs are formed in quasiperiodic optical potentials 5,6 .
For simplicity, we consider a one-dimensional (1D) analog to QC known as the Fibonacci chain, LSLLSLSLLSLLS::: (see QC in Fig. 1c) 7 , where L and S are long and short segments with the ratio L=S equal to the golden mean τ ð1 þ ffiffi ffi 5 p Þ=2. This chain looks to have no order at a glance, but it has the perfect order as understood from the fact that it was created by successively applying the self-generation rule, L ! LS and S ! L, onto the first generation sequence, L, as demonstrated below, It may be noticed that n-th generation sequence is produced by placing ðn À 2Þ-th one on the right-hand side of ðn À 1Þ-th one. Then, the total number of the L and S segments of the n-th generation, F n , follows the relation, This recurrence relation with F 1 ¼ 1 and F 2 ¼ 1 gives the Fibonacci sequence, 1, 1, 2, 3, 5, 8, ... A series of the successive Fibonacci number ratio, F nÀ1 =F nÀ2 , approximates the golden ratio; 1/1, 2/1, 3/2, 5/3,..., lim n!1 F nÀ1 =F nÀ2 ¼ τð¼ 1:6180:::Þ. There is an actual material that corresponds to each rational ratio and is called approximant crystals (ACs). Examples are shown in Fig. 1c; 1/1AC is a periodic crystal consisting of the unit cell LS, 2/1AC consisting of LSL, and so on. In F nÀ1 =F nÀ2 AC, F nÀ1 and F nÀ2 indicate the number of L and S segments contained in the unit cell, respectively. (In the 3D case, for example, 1/1AC denotes cubic 1/1-1/1-1/1AC.) This means that the unit cell size of AC increases with the order of the rational approximant.
Reflecting such the unique geometry, QC is expected to have an electronic state called critical state that is neither extended nor localized. The existence of extended eigenstates in periodic crystals is a consequence of Bloch's theorem, whereas in random systems, strong disorder can lead to the formation of localized eigenstates, i.e., Anderson localization, which occurs due to the interference effect between propagating and backwards scattered waves. In QCs, critical eigenstates emerge as a result of the competition between the broken translational invariance and the self-similarity of quasiperiodic structure 8 . Besides extensive studies, the electronic state of QCs is veiled in mystery 9 . For example, an electronic long-range-ordered states is not established yet although it was observed in ACs 10,11 ; to the best of our knowledge, there is no QC presenting the convincing evidence for bulk superconductivity [12][13][14] , i.e., zero resistivity, Meissner effect, heat capacity jump, and the fivefold rotational symmetry as well. (In ref. 14 , Mg 3 Zn 3 Al 2 was considered as a superconducting QC, but it seems to be AC according to the phase diagram given in ref. 15 and the present study, see below.) It is therefore interesting to discover superconductivity in QC. It is also interesting to examine whether the emerging superconductivity shows weak-coupling, spatially extended Cooper pairs or strong-coupling, local pairs (reflecting the critical state).
Here, we study the Al-Zn-Mg system as a test material owing to two reasons: First, it contains both QC 15,16 and AC phases 15,17,18 , and second, the AC phase exhibits superconductivity 14 . We show that bulk superconductivity emerges at T c ffi 0:05 K in the Al-Zn-Mg QC, implying that it is not only the first superconducting QC but also the first QC exhibiting the electronic long-range order. We also show that temperature dependences of the thermodynamic properties and the upper critical filed are understood within the weak-coupling framework of superconductivity, suggesting the formation of spatially extended pairs.

Results
Sample characterization. Samples prepared here are Al-Zn-Mgbased QC, 2/1AC, and 1/1ACs, which are summarized in the ternary phase diagram (Fig. 2). As reported in ref. 15 , the 1/1ACs have a wide composition range. In this paper, each 1/1AC sample with different composition is identified using the alphabetical character, e.g., 1/1AC_A. The 1/1AC_G is a mother alloy of the QC and has almost the same composition as the 2/1AC. Note that the alloy Mg 3 Zn 3 Al 2 mentioned above is close to the 1/1AC_E sample.
The structure of the obtained samples was studied by X-ray and electron diffraction method. The lattice constant a of the 1/1AC samples is illustrated in Fig. 3a as a function of Al content. We note that a decreases almost linearly with the Al content.
For the QC, the following indexing scheme of the reflection vector g is used in this paper; Here, the set of integers, m i , represents reflection index. The vectors e ik have a length equal to 1= ffiffi ffi 2 p , and they are parallel to the lines connecting the center of an icosahedron and the surrounding six vertices as in Fig. 6 of ref. 19 . The lattice parameter a 6D of the 6D hypercubic lattice may be related to the edge length a R of the rhombohedral cells of the 3D Penrose tiling as follows, In the following, we focus on the QC, the 2/1AC, and the 1/1AC_G (the mother alloy of the QC). The representative X-ray diffraction patterns of them are displayed in Fig. 3b, confirming almost the single phase. The diffraction peaks of the QC were indexed using the 6D lattice parameter, a 6D ¼ 0:7308 ± 0:0001 nm. Absence of any extinction condition indicates P-type of icosahedral QC. For the 2/1AC, the peaks are labeled using the lattice parameter, a 2=1 ¼ 2:3006 ± 0:0004 nm, indicating P-type cubic phase. The intensity is strong for the 850 reflection (d ¼ 0:244 nm) and the 583 reflection (d ¼ 0:232 nm). Note that these indices are combination of successive Fibonacci numbers. For the 1/1AC_G, similar results are obtained: The peaks were indexed with the lattice parameter, a 1/1 = 1.4195±0.0003 nm, for I-type cubic crystal. Note that the 530 reflection with the spacing d ¼ 0:243 nm and the 352 reflection with d ¼ 0:230 nm have a strong intensity. These indices are combination of the successive Fibonacci numbers again, and they correspond to 211111 and 221001 reflections of the QC, respectively.
The unit cell sizes obtained above satisfy the following equation 20,21 , This ensures our assignment of the QC and the AC samples.
Using the X-ray diffraction peaks around 2θ ¼ 65 , we estimated the correlation length as 47, 85, and 28 nm for the QC, the 2/1AC, and the 1/1AC_G, respectively. Comparison between them suggests that the 1/1AC_G is meta-stable at the composition Al 14.9 Mg 44.1 Zn 41.0 . For further comparison, we evaluated the correlation length of the 1/1AC_A as more than 65 nm, which is twice the 1/1AC_G value. This difference in the sample quality would yield the sample dependence in the physical properties among the different ACs (Supplementary Figure 2).
Electron diffraction patterns of the QC, the 2/1AC, and the 1/1AC_G are demonstrated in Fig. 3c-g. Figure 3c displays a fivefold diffraction pattern of symmetry m35 of the QC. Indices of reflections A and B are 122010 and 221001, respectively. Magnified image (Fig. 3d) including reflection B shows deviation from the exact regular pentagon for weaker reflections, indicating the presence of linear phason strain 22 . Figure 3e shows a twofold diffraction pattern of the QC. Indices of reflections A and C are 122010 and 121111, respectively. The τ 3 -scaling agrees with Ptype of icosahedral QC. Figure 3f and g shows diffraction patterns of the 2/1AC and the 1/1AC_G, respectively, with the incident beam along each [001] direction. The 2/1 and 1/1ACs show no fourfold but twofold axis. In Fig. 3f, indices of reflections D and E are 10 00 and 850 of 2/1AC, respectively. We observe the reflection condition that h is even for h00 and hk0 reflections. The 0k0 reflections with odd k should disappear following this reflection condition, but they are actually observed due to multiple diffraction effects. This observation is consistent with the space group Pa3 proposed for the 2/1AC 17 . In Fig. 3g, reflections F and G correspond to 600 and 530 reflections of the 1/1AC, respectively. Note reflection condition of h þ k þ l ¼ even for hkl reflection, which is consistent with the reported space group Im3 18 .
Electrical resistivity. Figure 4a shows the electrical resistivity normalized by the resistivity at T ¼ 280 K, ρ=ρ 280 K , as a function of temperature T in a logarithmic scale. Three points are to be noted. First, all the materials studied here show zero resistivity. Second, ρ 280 K of the QC and the 2/1AC amounts to~150 μΩ cm, greater than that of all the 1/1AC samples (inset of Fig. 4a). Third, while all the 1/1AC samples present the metallic behavior, the QC and the 2/1AC show the negative temperature coefficient of resistivity, dρ=dT<0 (Fig. 4b).
The normal state conductivity of QCs has been sometimes discussed using the concept of the Anderson localization 9 . For the present case, it remains open if the second and third points mentioned above show a precursor of the electron localization in the QC and the 2/1AC. This should be examined in the future by virtue of phason-strain-free samples; the present sample contains a linear phason strain as mentioned above.
Specific heat in normal state. The temperature dependences of the specific heat CðTÞ in the normal state of the QC, the 2/1AC, and the representative 1/1AC samples are shown in Fig. 5a in the form of C=T vs T 2 . Using the relation, C T ¼ γ þ βT 2 , we obtain the coefficients γ and β for each sample. The Debye temperature Θ D is deduced from β and plotted in Fig. 5b as a function of Al content, in good agreement with the previous report 15 . We confirm that Θ D is almost independent of Al content. For the Alcontent dependence of the electronic-specific heat coefficient γ, see below.
Relation between T c and 1=γ. Figure 6a shows the superconducting transition temperature T c defined by zero resistivity as a function of Al content. Note that the zero resistivity corresponds to the heat capacity jump (see below) and hence shows the bulk transition of superconductivity. As Al content is decreased, T c is monotonically decreased from ∼0.8 to ∼0.2 K, followed by the sudden drop down to~0.05 K at 15% Al content (corresponding to the QC, the 2/1AC, and the 1/1AC_G). Figure 6b shows the Al-content dependence of the electronicspecific heat coefficient γ deduced from Fig. 5a. We observe that γ monotonically decreases with Al content, suggesting that the density of states at the Fermi energy E F , D E F ð Þ, decreases with Al content. Note that γ slightly drops at 15% Al content, which is likely related to the electronic stabilization effect, i.e., the pseudogap formation due to the so-called Hume-Rothery mechanism 15 .
To see the relation between T c and γ, we plot ln T c vs 1=γ in Fig. 6c with Al content as an implicit parameter. We find that all the samples lie on the straight line within an experimental uncertainty. According to the BCS theory, T c is given as follows, Here, V is the effective electron-electron interaction with the weak-coupling condition jVD E F ð Þj ( 1. As Θ D is almost independent of Al content in the present system as mentioned above, Eq. (5) leads to the relationship, ln T c / 1=γ, if V is the same among the samples. This is just observed here, meaning that the effective interaction V remains attractive and unchanged in magnitude under variation of the atomic arrangement from the AC to the QC and T c is fully determined by D E F ð Þ.
Bulk transition of superconductivity in QC. Let us focus on the superconducting transition of the QC. (See Supplementary Figure 2 for the AC samples.) At T c marked by the resistivity drop (Fig. 7a), the real part of the ac magnetic susceptibility (χ′) becomes negative (Fig. 7b), signaling the shielding effect associated with the zero resistivity. Upon cooling the sample through T c under an external magnetic field, the dc magnetization M becomes diamagnetic (Fig. 7c), indicating the exclusion of the magnetic flux due to the Meissner effect. As seen in Fig. 7d, the specific heat divided by temperature C e =T shows the large jump ( δC e =T c $ 1:2γ) at T c , where C e denotes the electronic part of specific heat, obtained by subtracting the lattice contribution from the measured specific heat, and δC e indicates the jump height of C e . This indicates that almost all mobile electrons in the sample participate in the superconductivity. These provide convincing evidence for the emergence of bulk superconductivity in the QC. In Fig. 8, we show the normalized specific heat C e =γT of the QC and the 1/1AC_A as a function of the reduced temperature t ¼ T=T c . (The QC sample presented here is different from that shown in Fig. 7d.) We observe that the data of the QC and the 1/1AC are in good agreement with each other. Note that both the Superconducting critical field. The magnetic field dependence of the electrical resistivity ρðHÞ is demonstrated in Fig. 9. The zero resistivity defines the upper critical field H c2 shown in the inset of Fig. 10. Note that the 1/1AC_F has a several times larger H c2 than Al metal, while it has a several times lower T c . This excludes the possibility that the superconductivity might arise from Al-derived impurity phase. Combining the relations, κ ¼ H c2 ð0Þ= ffiffi ffi 2 p H c 0 ð Þ and H c 0 ð Þ ¼ T c ffiffiffiffiffiffiffiffiffiffi ffi 5:94γ p (where κ is the so-called GL parameter, H c2 ð0Þ and H c 0 ð Þ are the upper and the thermodynamic critical fields extrapolated to zero temperature, respectively), we evaluate κ as 136, 128, and 337 for the QC, the 2/1AC, and the 1/1AC_F, respectively. These values confirm that the present system is a type-II superconductor, in which the magnetic field penetrates the sample. The coherence length ξ 0 ð Þ was also evaluated from the relation H c2 0 ð Þ ¼ ϕ 0 =2πξ 0 ð Þ 2 (where ϕ 0 is the flux quantum) as ξ 0 ð Þ $139, 143, and 83 nm for the QC, the 2/1AC, and the 1/1AC_F, respectively.
The reduced upper critical field is defined as h ¼ ÀH c2 =ðT c dH c2 =dTj T¼Tc Þ, and is plotted in Fig. 10 as a function of the reduced temperature (t ¼ T=T c ). We compare hðtÞ with Werthamer-Helfand-Hohenberg (WHH) theory 23 , which takes into account of electron mean free path (l), spin-orbit scattering, and spin paramagnetism. The experimental results are in good agreement with the theory (solid line) for the case of no spin paramagnetic or spin-orbit effects and in the dirty limit (ξ 0 ð Þ ) l), in which scattering from physical and chemical impurities is large compared with the superconducting energy gap. This dirty-limit superconductivity seems compatible with the large coherence length estimated above and the large residual resistivity (i.e., small mean path) shown in Fig. 4. On the other hand, the present system is distinguished from some dirty Here, C e is the electronic part of the specific heat. The solid line denotes the weak-coupling BCS theory. Note that both results of the QC and 1/1AC samples are compatible with the weak-coupling theory energy, i.e., VD E F ð Þ>0 in Eq. (5). For the present QC case, two points are to be noted. First, D E F ð Þ is reduced presumably due to the pseudogap formation, and T c is much smaller than that of the ACs but remains finite. (The very low T c due to the pseudogap may explain why superconductivity was hardly observed in QCs.) This situation resembles that in superconductors in which charge-density-wave (CDW) states coexist; the Cooper pairing and the CDW instabilities compete for the Fermi surface and so the presence of the CDW depresses T c 25 . Second, the fact that V remains intact in the QC leads to the following discussion: The electron-electron interaction is expressed as V ¼ V a À V C , where V a is the attractive pairing interaction (mediated by phonons in conventional BCS superconductors) and V C is the effective Coulomb repulsion. If the critical eigenstates of QCs would lead to the localization effect and, as a result, cause slow diffusion of electrons, then V C could be enhanced and V would be reduced 26,27 . The absence of such the reduction in V implies that the critical eigenstates would not have a dominant role in the superconductivity of the present QC.
In this study, we found no difference between the Al-Zn-Mg QC and other weak-coupling superconductors. According to a theoretical study by Sakai et al. 28 , however, the Cooper pairs in the Penrose lattice are unconventional because the lack of the translational symmetry does not allow the conventional Cooper pairing formed at the opposite Fermi momenta, k and Àk. It would be challenging to detect the fractal superconducting order parameter as predicted by the theory. We hope that the present study stimulates a further work to reveal this new type of superconductivity.

Methods
Sample preparation. The 1/1AC samples were prepared by induction melting of appropriate amounts of constituent elements, 99.99% Al, 99.9% Mg, and 99.99% Zn, in a boron-nitride crucible under Ar atmosphere 15 . Some of them were annealed at 300°C for 6 h or at 360°C for 5 h. The mother alloys of the QC (Al 14.9 Mg 44.1 Zn 41.0 ) and the 2/1AC (Al 14.9 Mg 43.0 Zn 42.1 ) were first prepared by induction melting of the constituent elements. Then, by melt spinning of each mother alloy 15 , the ribbon specimens were fabricated. Finally, the QC samples were obtained by sintering the ribbons at 300°C and at 50 MPa for 1 h using a spark plasma sintering apparatus, whereas the 2/1AC samples were obtained by sintering the ribbons at the same conditions as the above and subsequently annealing the sintered ribbons at 300°C for 5 h. Some of the QC samples were annealed at 360°C for 5 h, whose structure was confirmed to be kept in the QC.
Sample characterization. The composition of the obtained samples was analyzed by using inductively coupled plasma (ICP) spectroscopy and scanning electron microscope (SEM). For the ICP, the analyzed composition agreed well with the nominal one within the error <2%, and no segregation was detected for the SEM within the experimental accuracy (Supplementary Figure 1).
Selected-area electron diffraction patterns were observed using a JEOL JEM-200CS microscope with a double tilting stage at the acceleration voltage 200 kV. The alloy specimens were crushed into fragments using an agate mortar and pestle, and transferred on a micro-grid mesh for the electron microscopic observation.
X-ray diffraction patterns were obtained using a RIGAKU IIB diffractometer. Lattice parameters of the QC and the ACs were determined from angles of Bragg reflections, θ, using the extrapolation method: Least square fitting and extrapolation to θ ¼ 90°were carried out by assuming linear relationship between calculated lattice parameters and the following equation values, cos 2 θ sin θ þ cos 2 θ θ :  Correlation length, L (nm), was determined using the following relation, Here, λ (nm) and Δθ (rad) denote wave length of X-rays and peak width (full width at half maxima), respectively. In this study, the following three reflections, 332002, 8 13 5, and 583 reflections, located approximately at 2θ ¼ 64:8°, were used for the QC, the 2/1 and 1/1ACs, respectively. To estimate Δθ, the peaks were decomposed into two parts originating from Cu-Kα 1 (λ ¼ 0:15405 nm) and Kα 2 by assuming pseudo-Voigt function for each peak shape.
Physical properties measurements. The physical properties were measured using one 3 He refrigerator and four 3 He/ 4 He dilution refrigerators (each having a different base temperature) installed at Nagoya and Tohoku Universities. Different measurement techniques were taken depending on the temperature region measured: for the electrical resistivity, a four-terminal dc or ac method was taken; for the ac magnetic susceptibility, the mutual inductance method or a SQUID magnetometer; for the heat capacity, the quasi-adiabatic heat-pulse, or relaxation method. The dc magnetization measurement was done using a SQUID magnetometer.
Data availability. The data that support the findings of this study are available from the corresponding author (kensho@cc.nagoya-u.ac.jp) upon request.