Girsh Blumberg (Rutgers): Raman spectroscopy of multiband superconductors with competing order parameters

Girsh started by a remark that his talk will be devoted to the many-body aspects of multiband superconductivity as seen from Raman spectroscopy.

First example given is a Raman study of layered system, NbSe_2, by Sooryakumar & Klein, PRL 45, 660 (1980), the material with coexisting CDW (T_CDW ~ 9K) and (multiband) superconductivity (T_c ~ 2K). The first challenge for Raman spectroscopy was to keep the material below the sc transition temperature as laser heats the sample [remark about the Burch's talk earlier today].
On the theory side the story began with the study by Abrikosov and Falkovskii in 1961.

What is interesting on the side of multiband superconductors? Following Littlewood and Varma in 1982 there are two possible modes in a usual superconductor: amplitude mode (amplitudon) and the phase (Bogolyubov-Anderson ) mode. In a multiband SC there is an additional Legget's mode associated with the phase difference of the sc gaps on different bands.

Next Girsh has reviewed the classical work by Klein and Dierker, PRB 1984. In the normal state
the only electronic excitations Raman spectroscopy is able to see is a particle-hole continuum with a small (almost zero) momentum transfer. In a superconductor due to renormalization of the electron bands the continuum will be gaped up to a twice of the sc gap. Observed by Hackl (at a time graduate student) in V_3Si for the first time. HOWEVER: this is all for non-interacting electrons: in case the residual interaction is involved the peak at 2\Delta could be shifted to the lower energies and whether it happens or not depends on the symmetry of the OP, Raman vertices and so on. Special example is a long-range Coulomb interaction. The excitation driven by light are electron charge densities which should be screened by the long-range Coulomb interaction especially in the fully symmetric A_1g channel.

Then he moved to cuprates, and has shown experimental examples of Raman scattering in the various (B_2g, B_1g, and A_1g) symmetry channels. Due to the fact that Raman vertices are momentum dependent and have maximums and minimums at the various parts of the Fermi surface you realize that B_1g probes mostly quasiparticles around the M point of the BZ, while in B_2g you taking mostly the excitations around the diagonal of the BZ. Naturally if the gap is d-wave the largest effect comes in the B_1g channel [few curved from YBCO]. Then he moved to the story of interaction and reminded about the work by Chubukov and Deveraux on the final-state interaction effects in the B_1g channel (below T_c). Comments by Chubukov: first work is with Girsh. Comment by Nevidomskii yielded the understanding among the audience that what we hear today is only the electronic Raman scattering. [no two magnon, no phonons]. Hirschfied pointed out that in the data the peak is hardly distinguishable from usual 2\Delta peak, Chubukov replied that the story actually is that the peak in Raman is BELOW 2\Delta and the latter is measured by other techniques and this is what effect of the interaction is: shifting the peak to energies below 2\Delta.

Then Girsh turned to the electron-doped cuprates (PCCO) and the story of electron and hole pockets there. Blogger is busy typing but notices that the story itself is quite interesting and for non-expert probably heavy to follow with one slide that Girsh has shown. Anyway if you accept that there are pockets you understand the asymmetry of Raman spectroscopy in electron and hole-doped cuprate superconductors.

Ok now we are back at the Legget's mode as Girsh comes up with the paper by Klein, PRB 2010 where he
analyzes Raman in MgB_2 (by now classical two-band superconductor). By looking in E_2g symmetry we notice two separate 2\Delta structures associated with pair-breaking of the correspodning gaps [temperature dependece, coupling to phonons]. Now if you look at the A_1g you see an extra (unexpected) feature at 9.4 meV which, as Girsh argues, is a collective mode effect. To understand this you look into the theory of a two-band superconductors:
For two order parameters the phases of the order parameters are arbitrary but if there is an interaction you can have in phase locking for an attractive interaction (V_interband>0) and out-of-phase locking of the phases for the repulsive interaction (V_interband<0). Klein pointed out that in MgB_2 the interband interaction is large and positive. The former shifts the collective mode associated with this phase locking to higher energies (into continuum) which yields damping.

Now iron-based supercondctors (from 2 to 5 bands). The idea about collective mode goes back to Chubukov, Eremin, Korshunov, PRB79 (2009) [thanks Girsh!]. They story is that if the symmetry is extended s-wave and these are electron and hole pockets involved then the effect of the final state interaction will be to create a collective mode below 2\Delta. Now the apparent contradiction [after discussion of Chubukov, Hirshfield, Fernandes, and Nevidmoskii, blogger also liked to contribute but had to continue typing] was that the collective modes Girsh has introduced earlier had nothing to do with the collective mode in B_1g in cuprates and in the A_1g channel of iron-based superconductors which he discussed now. These are all density-density excitations (excitons) shifted to lower energies due to interference of the the gap and Raman vertices. So these are neither Leggett's or amplitudon modes.

Finally in the last few minutes Girsh comes back to NbSe_2. He discussed some STM and ARPES data by Hanaguri et al., and Borisenko et al., respectively. What about Raman? Well, it seems that you see 2\Delta features only (sc and CDW). However, it is bit controversial because Girsh suggests that it might be an amplitude mode.

Questions: 1 ) blogger: why one should believe that the intreband intreaction in NbSe_2 is repulsive, Answer: the system is complicated not that much is known about the system.
2) questions raised by Nevidmoskii produced some discussion which has moved to a lunch and basically concerned the symmetry representation of the various channels in the B1g, B2g and A1g symmetry representations for iron-based superconductors. It seems that to correctly interprete the data of Muschel et al you have to work in the unfolded BZ.

Yoshihiro Iwasa (Tohoku U): Electric field induced superconductivity with electric double layer transistors

Outline:

• electric field induced SC
• double layer Transistor
• oxides
• increase of carrier density
• layered ZrNCl
• summary

Intro:

charge accumulation in capacitor lead to MOSFET by replacement of one electrode structure, however, difficult to make in the past

at the same time (Glover 1960) electric field control of SC was investigated (In)

field effect on high-Tc cuprates (YBCO): by applying gate voltage increase of Tc

dream: gate control of localization-SC transition (e.g. a-Bi, Parendo PRL 2005)

best would be SC in an insulator by gating

conventional FET with weak electric field ~1 MV/cm, 10^13 cm^-2 which is too little, therefore attempt to use liquid gate (ionic), this leads to charge accumulation devices like Li battery or electric double layer capacitor

comparison of capacitance: double layer capacitor b/w electrolytic capacitor and battery, in double layer capacitor voltage drop right at the electrode

application of EDLT to organic conductors: sharp raise of conductance --> low operation voltage, but conductance in order of 4 microS therefore switch to inorganics


made of ZnO and patterned, ZnO shows large increase of sheet conductance at about 1.7 V, carrier densities are several 10^13 cm^-2, the capacitance is 7.8 microF/cm^2, e. i. half that as on Au

low T measurements;

crossover from insulator to metallic behavior at 0.7 V
now STO/PEO(KClO4) is being used, current two order of magnitude larger, insulator to metal transition at around a few volts

electric field induced SC was then found at V_g = 3 V, first field induced SC without doping, critical magnetic field about 30 mT, H_c2 is smaller than in the bulk due to interface which is not working as a pinning center

increase of carrier density:

uniqueness of STO:
STO is lowest carrier density SC, also atomically flat surfaces can readily be made

higher carrier density in ionic liquids: polymer electrolyte (solvent + salt) but large solvent molecules are bad, therefore solvent is to be removed --> melting at RT (organic material)

now accumulated carrier density is in range of 10^14 cm^-2, gate voltages of 0.5 V needed at room temperature, gate V goes up with T --> enhanced electric charging at low temperature, that should be advantageous for SC



Device:


ZrNCl:

discovered about ten years ago, layered material, cleaveable and interesting phase diagram

very small specific heat capacity if compared to other SC but relatively high Tc of ~15 K (goes down with Li-doping, pairing interaction increases which is strange, magnetic fluctuation are present but it is not clear where they come from, crossover from isotropic gap to an anisotropic one

try to explain these strange properties by a fluctuation exchange approx. theory, afm fluctuation develops causing d pairing even in doped band insulator







device fabrication:

exfoliating single crystals, electrodes by e-beam lithography, TiAu electrodes, single crystal size some 10 micrometer,

sigma kicks in at around 1,5 V at 220 K, goes up to 1.5 mS, by increasing the gate voltage remarkable decrease of resistivity, at about 4 V SC is found below ~14 K, however it is not clear how much volume of the sample turns SC

doping study: Tc goes down to below 12 K but there seems to be a maximum at ~0.04 as shown by recent low-doping studies

novel materials:

also, KTaO_3 turns SCbelow 0.04 K

• new tunable 2D systems
• challenges: increase Tc, discover new SC
• new states at interface which are inaccessible by conventional chemistry

Ken BURCH (U. Toronto): Tuning Materials with Mechanical Exfoliation

In order to tune properties - how to tune carrier densities without affecting purity, homogeneity etc.

OUTLINE:

• Introduction
  
• Method: exfoliation. A very expensive glove-box is (almost) all one needs.

• Materials: BSCCO, Bi2Se3. Using spectroscopy (Raman) in order to characterize the tickness of the layers.

E.g. in BSCCO. Characteristic features in the Raman spectrum: a low and a higher energy peaks. Changing the doping level enhances the intensity of the higher-energy (3-phonon) peak.

• Future directions

I. Introduction.
Disorder vs. doping.
E.g. STM work in BSCCO-2212 (McElroy, Science 309, 1048 (2005)) shows a lot of disorder on the surface (and presumably the bulk too). One would like to tune the doping without affecting the disorder. Exfoliation technique is promising.

Spectroscopy is a powerful tool to study materials, e.g.:
- ps-gap in the cuprates. Optical studies (Basov and Timusk, RMP 77, 721 (2005)) were one of the first to show the existence of the ps-gap. Also Fermi-arcs were observed in ARPES (Tanaka, Science 314, 1910 (2006)). Raman studies (Blumberg, Science'1997).
- Fe-pnictides - a new exciting direction.

Interfaces and surfaces (epitaxially grown)
- Ohmoto et al, thin films of BaTiO3/ Nature ()
- ferroelectrics

Field-effect transistor.
Consider SiO2 grown on top of Si:P.
There are few examples of combined thin layer materials (e.g. LaAlO3/LaVO3 etc), but material and growth issues are daunting. Lattice mismatch is important and prevents one from growing arbitrary combination of the film/substrate.
The exfoliation technique, on the other hand, is free from this drawback - the same thin film can be exfoliated onto a number of different substrates.

II. A new way: exfoliation technique
Pioneered by Andrew Geim for graphene. A.K.A. "skotch-tape" technique

- graphene
E.g. Tony Heinz's group showed that if grown on MICA, the graphene layer is very flat, unlike if grown on e.g. silica.
(The blogger did not know what MICA was - as far as I understood (please correct in incorrect) it is a substrate typically used in AFM for calibration. The surface of it is covered with Na aroms, leaving an atomically flat surface underneeth it)

- topological insulators
See Peng, Nat. Mater 9, 225 (2009); D. Hsieh, Science (2009)
e.g. Bi2Se3, Bi2Te3
Since MICA is transparent, the speaker used it for exfoliation, and then measured Raman on it.

- NbSe2
Raman A1g mode shifts depending on the thickness of the layers.

- MoS2
A. Splendiani.., Nano Lett. 4, 1271 (2010)

III. Results on BSCCO-2212
Bi2Sr2Ca0.6Dy0.4Cu2O(8+x)
x = 0.3, 0.4, in the ps-gap regime

Depending on the film thickness, the colour of the crystal is changing. Where does it come from? Chemistry is unchanged! Turns out - interference effect by reflection from different layers. Plot so-called contrast depending on the layer thickness - convincing.

Lab Tour: 9 T magnet, ellipsometer, Raman scattering ...

History of Raman: sunlight from a telescope, passed through a polarizer, filters to a single wave-length, and scatters it from the sample. The shift in frequency (Raman shift) is what measured.

Sharp features in Raman - 2-phonon joint density of states.
Broad features: two-magnon excitations. Review: T. Deveraux and R. Hackl RMP 79, 175 (2007)
E(1 magnon) = 2J;
E(2 magnon)=3J - most pronounced signal, since one breaks 6 spin-spin bonds by flipping two adjacent spins, paying energy cost J/2 for each.
Note: Other frequencies are possible (e.g. E(2 magnon) = 4.5 J).

The 2-magnon excitations have definite selection rules. E.g. B1g excitation corresponds to XX or X'X' polarization. The intensity of the Raman signal from thin films (exfoliated direcltly on the SiO2 substrate) greatly enhances compared to the bulk (See data in: Sandilands, PRB (2010)).

NOTE: Raman signal, even from the bulk, always has interference correction from reflection off different layers. (See Y. Wang, APL 92, 043121 (2008) ).

So what is the reason for much enhanced intensity of the B1g 2-magnon peak?
It turns out - this is the effect of the changed doping level. In fact, the bulk data show that reducing concentration of holes (under-doping) hardens the 2-magnon peak (i.e. shifts it to higher frequency). This is precisely what we observe in our exfoliated thin film flakes.

SUMMARY:

• using exfoliation technique allows to create very thin films without worrying about the interaction with the substrate, lattice mismatch etc.
• optical studies are a very powerful method
  
• thinning out the BSCCO films seems to result in increasing under-doping, judging from the 2-magnon Raman peak frequency.

Questions:
Q: P. Hirschfeld. Presumably, the reason for under-doping in exfoliated films is that you lose O atoms from the surface. However, don't you think that only few top layers would be affected?
A: It may well be true. Our data show however that the effective doping level, as measured by Raman 2-magnon peak position, reduces. We are not quite sure at present, due to what microscopic mechanism.

Q: H. Alloul. Are data taken at room temperature
A: Yes. The low-temperature data can be different

Q: A. Chubukov. Were all the measurements done for one doping level?
A: Yes, so far we started from overdoped O-concentrations. We would like in future to do exfoliation starting from optimally doped samples. We have also experimented with different Dy doping (not shown in this talk).

Q: G. Blumberg.
What was the laser power
A: a few miliwatts.
Blumberg's comment: concentrating this much power on a tiny surface area would lead to heating of the sample, and hence oxygen diffusion off the surface.

Q: P. Coleman: what do you know about the dependence of exfoliation on temperature?
A: we heat the substrate before exfoliation. We haven't yet done the exfoliation at low temperatures.
Q: Is the size of the flakes sufficient to attach leads?
A: The flakes are too small. It would be great to be able to make them larger.

Q: A. Schofield. What are the prospects of this exfoliation technique?
A: Exploring the phase diagram of e.g. cuprates, without affecting the purity of the crystals.

Q: Y. Grin. What do you know about the real structure of the spacer? Can you do TEM?
A: TEM proves hard. We do AFM, showing very flat surfaces. We'd like to do other structural probes in the future.

Q: [someone]. Can one see a gradual change in the Raman B1g position from bulk to thin film, by gradually increasing the film thinkness?
A: No, it's very hard to control the film thickness during exfoliation process. Somewhat of a black art.

Q: P. Hirschfeld. How important is lattice mismatch with the substrate? E.g. other studies on epitaxially grown films?
A: We hope that we are not stretching the film, but we cannot really be sure. We tried different substrates to try to answer this question.

Henri Alloul (Orsay): NMR studies of the pressure induced Mott transition to superconductivity in the two phases of Cs3C60

Henri started by asking the audience whether they are familiar with the fact that fullerenes are superconducting and also strongly correlated systems [majority knew, by the way]

He also pointed out the major players in the field in a past: P.W Anderson and B. Batlogg

Highlight of the field: new results on the superconductivity induced by pressure, Takabayashi et
al, Science 2009, A. Ganin et al., Nature Materials, Aprl 2008

Outline of the talk:
- Introduction: A3C60 and their superconductivity
- electronic corrleations and Jahn-Teller Distotions
- Expanded magnetic moments
- Conclusion


He introduced the crystal structure (cubic and fcc - bipartite) of fullerides. Electronic structure from LDA: from molecular levels to the Bloch states [transfer integrals are weak W~ 0.5 eV] t_{1u} bands are filled by introducing dopants (alkali ions)

Exp. fact: Tc depends linearly on the lattice constant, a. It was originally interpreted as a sign that BCS formula Tc~ \hbar \omega_D exp[-1/(V N_0)] works well. Thus it was concluded at a time that it is a phonon-mediated superconductor [in 1991!]

1995: the story is not so simple - different behavior of the slopes Tc vs a [lattice constant] for Na2AC60 and A3C60 - signs of the correlations.

Furthermore, for A_nC_60 it was found that they have different (from metallic to insulating) behaviors. He gave two examples: A_4 C_60 (bct structure) and N2C60 (cubic). Both show small spin gap from NMR and large charge gap from optics - Mott insulators?!

The reason: large Coulomb interaction U, two electrons on a ball costs an energy U plus there is a Jahn-Teller effect [deformation of the molecule]. Remarkable result is that for n=2 and n=4
there is a larger energy gain per electron [these are results from the molecular structure calculations by Tosatti]. Especially there is additional U_{eff} which arises due to Jahn-Teller distortion and adds to the usual U for even n (in A_nC_60) and U-U_{eff} for odd n. It gives Mott insultor for n=2 and n=4.

Nevidomskii: why it is not simply a band inslutor: Answer: the reason is that n=2 and n=4 have a different crystal structure, it cannot be explained on the level of band insulator.

The rest of the talk was about odd value of n in A_nC_60

Special case: CsC_60 (A_1C_60): - Mott insulator. The reasoning: from a high_T_c cubic phase phase - to a polymer phase at 200K - and finally dimer phase at 77 K (all from NMR). Experimental justification comes from NMR which sees 3 different nuclei sites with different electronic surrounding. Studying the intensity of NMR you realize the proportional compositions of the phases. By doing that you find 12% of sites in a spin singlet state.

Then he moved to the A_nC_60 series with n=3: here the most intriguing perspective is a search for the Mott insulator (originally studied in 90's K_3C_60 is not a Mott insulator). The idea is to take a larger alkali ionic radius (going from Li to Cs). Success by chemists: Cs_3C_60 has been recently prepared. Here you do find the AF Mott state and overall the phase diagram as a function of pressure resembles many of those which are typical for Mott insulators: AF at small pressure, AF+SC at intermediate pressure (doping) and SC at larger doping with a dome like structure. The slight complication is that there is also a structural transition in these compounds: A15 structure for Cs_3 C_60 - you see a single Cs site with non-cubic local symmetry (MIT and SC part of the phase diagram); fcc (only Mott part) structure of Cs3C60 you find two Cs sites and the ratio 1:2. Difference allows for selective NMR experiments.

In the next few slides Henri has analyzed the magnetic dynamics and crystal structure by means of NMR in Cs3C60 in the fcc phase. Special emphasis was put on the enhanced magnetic fluctuations in the paramegnatuc phase. Upon pressure you find the transitions from fcc phase to A15 phase as well the transition from AF to SC phase. Important remark: Mott insulator to metal transition is not directly related to the crystal structure transition. (here the argument is that you do have MIT also in the A15 phase)

Superconductivity: definitely singlet superconductivity most likely s-wave (there is a Hebel-Slichter peak). But more measurements have to be taken.



Summary:
- fullerides are correlated
- original due to disorder, icosaedral symmetry of the soccer ball
- supercondctors near MIT
- static charge segregation in Cs1C60
- importance of the Jahn-Teller effect: different between odd and even n
- excellent possibility to study the multiorbital Mott transitions.


Questions: 1) blogger: is Jahn-Teller splitting is comparable to the bandwidth, Answer: there is no clear indication, calulations are done for the molecule.
2) Vojta: what is the role of possible spin frustration in the MIT Answer: not really known
Remark: coexistence region between SC and AFM is possibly an inhomogeniety effect
3) Nevidomskii: fcc phase can be a spin glass - any results for zero field. Answer: similar to the previous ones: the field just started, Remark from someone in the audience: there are data from muSR which indicates phase transition or something similar at 2K, origin is unclear.

Dirk K. Morr: Defects, Density of States and Differential Conductors in Heavy Fermion Materials

Dirk's work is done with Jeremy Figgins.

Dirk begins with the puzzle of the resistance minimum in metals (de Haas et al., Physica 1, 1115 [1934]) and in "pure" gold sees that minimum tuned with magnetic impurities. We get the history...Kondo, Wilson, large-N...: Then Dirk introduces the spectroscopic signature in dI/dV (V. Madhavan et al. Science 280, 567 [1998]) where the Fano lineshape is identified in tunnelling as a signature of the Kondo resonance developing. So much for the single impurity.
Now we start thinking about the Kondo lattice of magnetic atoms with a conduction fluid. Dirk shows us the bad actors of quantum criticality (Au doped CeCu6; and magnetic field tuned YbRh2Si2). Question: about the interplay of doping and quantum criticality and how close you can actually get to the QCP with a discrete parameter in CeCu6-xAu? Stefan Wirth comments about how it can be combined with pressure to get to the QCP. The signatures of the non-Fermi liquid physics that emerge in these QCPs: Resistance T^n where n \neq 2 and T log T specific heat. (Note for the students from Andrey: technically n<=1 is really required for a non-Fermi liquid). Dirk now is moving on to the notion that impurities can be a useful probe of unconventional/puzzling systems. Cuprates provide a case-in-point: eg impurities in the superconducting state and the induced resonant states and quasiparticle interference experiments can probe the d-wave superconducting state. What about measurements of STM tunnelling on the heavy fermion systems. There are (at least) three groups working on this: Seamus Davis', Ali Yazdani's and the Dresden group with Stefan Wirth whose data are being discussed at this meeting. Ken asks about the symmetry breaking implicit in the introduction of a surface: Dirk "parks" the question. So, Dirk's key questions are

• Can defects provide insight into the heavy fermion systems and

• Do defects discriminate between electronic and magnetic correlations

The idea is that Friedel oscillations may provide that insight.
There are two possible defects: removing a magnetic atom (Kondo hole), or replacing a magnetic atom by a non-magnetic one. How should this be described? There has been previous work (Schlottmann, Freytag, Vojta and others). Dirk's approach begins with a Hamiltonian: Its the usual Kondo-Heisenberg Hamiltonian characterized by J (Kondo), I (Heisenberg) and an additional U0 (potential term on the impurity sites). His approach is SU(N), representing the spins as fermions which will be treated in mean-field which can have a local character. Rather than explain the actual calculations we get the physical content of the mean fields:

1. There is a hybridizing field (which is the Kondo physics) mixing s electrons and the spin fermions
   
2. The magnetic bond variable (this is the Heisenberg physics which gives the fermions representing the spin become itinerent)
3. A local constrain to force nf=1 forbidding valence fluctuations

Piers asks a nasty question: how can you have hopping of the f fermions when nf=1. Dirk gives the usual answer that it is a constraint on average. Piers says that is not acceptable. Piers: You should say we have a gauge model description whereby the "f" fermion is really a composite object which when included the gauge fluctuation and the hybridization means that it is representing not simply the moment but the Kondo-ized gauge field composite.

[An interesting blackboard discussion emerges after the talk (PC and DKM): the point that PC is making is that if you start from a Hamiltonian then the f in the Vc^{\dagger}f hybridization term is not really the original fermion which represents the spin rigorously localized, but is a composite object which involves a gauge fixed choice (Anderson and Appelbaum). This is significant because tunnelling a physical electron into that f object is, in reality, tunnelling into a complex many-body state containing the physics of "co-tunnelling".]

What does an STM experiment measure in HF materials? Dirk's work (Figgins and Morr, PRL 104, 187202 [2010]) (and other people...) stresses that there are two possible paths that the tunnelling electron could take: into the magnetic ion or into the conduction band and these two routes lead to the asymmetry and the Fano shape of dI/dV and the ratio of the two processes tf/tc radically changes the shape. tf/tc=0 has one shape while tf/tc=0.08 is enough to completely reflect the symmetry of the dI/dV curve. Similarly inverting the band structure also inverts the shape of the line, so bandstructure matter. Matthias points out that there is another possible asymmetry coming from moving away from the particle-hole symmetric Anderson model which everyone uses. Andriy asks how the calculation is done: since with tf/tc=0 does this not mean that the Fano shape should be Lorentzian? Answer is that there is asymmetry in the original bandstructure (unhybridized) - but then Dirk sketches it on the board and it looks pretty symmetric at low energy scales. Blogger is not convinced here but is too busy typing to ask the obvious question...[But now I can discuss it. The answer Dirk should (in my opinion) have given is that tf=0 does not mean there is no Kondo effect - it just that you tunnel only into a c-electron. The bybridizing mean field is still there so there is a Kondo resonance which is asymmetric with respect to the chemical potential].]
Dirk then tries to explain this with pictures of how the shape changes with the various parameters in the theory for the case of Cobalt on Gold (111). Andrey asks about the small scales that appear in the plots (meV) when the bare scales are (1eV) and how they come about? Dirk says it is TK. Markus asks for further clarification of the definition of the tf and how the process happens (like how the hole left behind affects things). Piers says you cannot do SU(N) for S=3/2 cobalt since S=3/2 is a symmetric representation of SU(2) and SU(N) large N does antisymmetric representations. Then into a discussion of spin-orbit as to whether J=4 saves you, but PC says orbital physics is quenched...an impasse. Then Dirk compares his theory with some of the experiments.
Finally we move to some numerical studies of Kondo impurities in small cluster numerical studies (arXiv:1001.3875) to look for the perturbations in the electronic correlations. He looks at how the hybridizaton and bond variables get distorted in his numerics. The role of the conduction Fermi surface and the gives significant changes (with large anisotropies) to the shape of the oscillations in real space. Dirk now needs to relate the oscillations and distortions seen in his mean fields relates to things you measure in experiment. Under pressure from the chair to wind up we get a fast tour through adding non- magnetic impurities. And in closing an array of Kondo holes is looked at which drives a first order phase transition as the holes start to interfer. Open questions: out of time.
Henri: a comment - these ideas have been explored in the cuprates by Henri for 18 years - not with STM but with NMR. Why not use NMR to do this in heavy fermions? Response: you need a magnetic impurity, Henri no you don't think of Zn in the cuprates.
What is the physical reason for tf << tc? Correlations suppress tunnelling into the f-electron state. You can also have co-tunnelling where in effect a spin hops (as studied by Piers Coleman and colleagues).
Peter Hirschfeld: Do you really see such a large factor of 10 in the anisotropy in the real physical systems? DKM: It may be a consequence of an idealized band-structures.
Stefan: In reality replacing a magnetic ion with a non-magnetic one - it matters greatly which atom you mess around with and you get very subtle changes in the vicinity of quantum criticality.

Determined Blogger: Andy Schofield

Matthias Vojta: Kondo impurities in Graphene

Matthias began with a summary of Kondo physics.

Matthias turned to the quest for realistic models for magnetic impurities. eg. its not so simple to figure out the Kondo model for Fe in gold.  Not well known until very recently, because of the interplay of spin and orbital degrees of freedom. Is it spin 1/2, or two channel spin 1, three channel spin 3/2, four channel spin 2?

Alloul pointed out from the high temperature chi in the early days, it was known to be larger than S=1/2.  Matthias argued that it did not really become clear until recently - Costi et al, PRL 102, 056802 (2009). Best fit to experimental dephasing rate suggests S=3/2, three channels.

The dephasing rate grows from T^2 up to a broad plateaux, and the subtle different fits favour S=3/2 for Fe in Au.

Outline of talk:

1 Impurities in Graphene, Dirac fermions, STM expts, orbital physics of d electron impurities.

2. Review: pseudogap Kondo model.  Quantum phase transitions.  Critical field theories.

3. Pseudogap Kondo model with voltage bias. (You can tune from linear density of states to a finite density of states. Maximal electron hole asymmetry. Spectral functions.

Turned to Dirac Fermions in graphene. Two atoms per unit cell.  When you diagonalize the short-range hopping Hubbard model (U=0), you get two Dirac cones with Hamiltonian

H ~ vF (p-K).sigma_sublattice
E_k = v_F }| k - K|

Pauli matrix acts in "sublattice space".  There are two copies of this Dirac Hamiltonian. The Dirac cones are "topologically protected".  (Semenoff 1984, Haldane 1988).   The Fermi points are robust against next nearest neighbour hopping etc.

By gateing, you can tune the Fermi surface to go from a Fermi point to a Fermi surface.

Dos (E) \propto |(E-E_F)|   linear density of states.

Kondo effect in graphene: first observation

Manoharan group. STM shows the hexagonal structure, with a puckered, rippled surface. Schofield asked why there was a superstructure.  There was no obvious answer from the croud.  From dI/dV you can see the Co on the surface. Now you can see "blue dots" representing the Cobalt atoms. You can now see the dI/dV spectrum.  You see a peak on some cobalt atoms, on others you see a dip.  Can extract a width, or Kondo temperature of TK~ 15K.   These pictures corresponded to an effective gate voltage of 200mV.

The point is, there are two different locations of the Cobalt atom.  Site A corresponds to atom on top of a C atom (t-site, dip structure, pseudo-spin breaking) whereas site B (h-site) is in the middle of a hexagon (peak, pseudospin conserving).  Add a field, the structures split, proving that the peak is of magnetic origin.

What is the correct Kondo model for Co on the graphene sheet?   The symmetries are very important here. In the graphene you have band degeneracies - C3nu, C6nu - three and six fold degeneracies.

Orbital physics and spin orbit coupling crucial.  Models such as SO(4) Kondo model are possible.   Using Generalized Gradient Approximation + U, a first principles study. Spin resolved DOS for Co in center of Hexagon and above C.  Three orbitals E1 (dxz,dyz) E2 (dx2-y2,dxy), A1 (d3z^2-r^2).  h-site, spin 1/2, SOC lifts 4 fold degeneracy - SU(2) Kondo possible.  h-site, spin 1 (SOC stablizes singlet, no Kondo expected.).  t-site spin 3/2 in E1, E2, A1 that would lead to a two stage, small TK effect.
(Wehling et al, PRB 81, 115427 (2010)). DOS is strongly particle-hole asymmetric.  J~ 2eV, bandwith from t=2.8eV.

Peter Hirschfeld asked where the spin is localized.  Henri Alloul suggested that the GGA+U might not have enough correlations to locate the spin. The blogger thinks these methods are probably good enough to get the spin form factor.

So what happens for the Kondo effect in a non-magnetic host.  If the DOS vanishes at the Fermi level, there is no Kondo screening at small J_K.  (Fradkin and Withoff - though not referenced).  Two possibilities

Hard gap - first order transition upon varying J_K at T=0.

Pseudogap - continuos transition upon variation of J_K. Non-trivial finite T behavior arising from quantum critical point.  DOS ~ epsilon^r.   r >0 gives phase transition.  (d-wave, graphene r=1).

Pseudogap Kondo model - 

Two axes - Kondo axis J.  Particle-hole asymmetry V.

For small r < r* = 0.3748, get simple Fradkin-Withoff behavior. Jc ~ r.  Also an ASC, asymmetric strong coupling fixed point.  beta (j) = rj - j^2.

For r* < r < 1/2  a new fixed point appears at finite Vc and Jc.

r=1 is upper critical dimension. r=0 is lower critical dimension. Hyperscaling is obeyed for r<1.

Chubukov asks can you do an expansion in epsilon = r-1?  Matthias says yes - but to do it requires the Gaussian theory at r>1.  The answer is a level crossing between a doublet of single impurity and a singlet of a screened impurity.  Simple model with doublet of energy epsilon-0 hybridized to a singlet via an Anderson screening - it is a non-interacting pseudo-gap Anderson model.   Can now do an epsilon expansion.  Vojta Fritz PRB 70, 094502 (2004)

r<1 have a finite hybridization fixed point (Wilson Fisher fixed point).  r>1 have gaussian fixed point. Critical fixed point is maximally p-h asymmetric near r=1. Hybridization becomes irrelevant above r=1, relevant below r=1.

Pseudogap Kondo model with voltage bias Sofar, only neutral graphene.  Next, mu>0. Now the moment will ultimately be screened at low T.  But if the chemical potential is of order the TK, there will be critical physics.  Chemical potential provides a fan of NFL physics.  J=Jc, then predict TK = kappa * mu.

RG now done with chemical potential effect on flow equations.   The leading effect is that one drives the impurity to the  screened, or unscreened phase.  (mu <0 epsilon = -infinity screened; mu > 0, epsilon = + infinity).  Depends on sign of mu.

Ultimate results - TK as a function of gate voltage.  (Vojta, Fritz, Bulla EPL (2010)).

Conclusions

• Magnetic impurities in graphene. Kondo criticality possible.

• Critical theory is not of Landau Ginzburg Wilson. but intrinsically fermionic

• TK(mu) extreme asymmetry between electron and hole doping, not only near criticality resulting from structure of critical fixed point.

• Systematic measuremets of Co impurities as function of gate voltage required. 

Kenji Ishida: NMR in Pncitides Talk 4 August 2010 11:30 am

Main message: AF fluctuations with quantum critical character induce superconductivity in Fe-based systems.

Focus in talk was primarily on BaFe_2(As_1-xP_x)_2 (P-doped 122). This system is interesting and unusual because a) it shows transport properties suggestive of quantum critical behavior near optimal doping at P=0.33; b) the SC dome is created by “doping” with P, which is however isovalent to As; c) the superconducting state properties are strongly suggestive of line nodes. Here are resistivity measurements from Matsuda group quite reminiscent of cuprates showing relatively large range linear fit to resistivity at x=0.33, and deviations away from this doping.

P-doping affords the possibility for NMR to do very clean experiment, since ^31P nucleus is spin-1/2 & eliminates quadrupole contribution to signal. At first glance the symmetric position of the As or P above Fe plane might make you think it would not be sensitive to AF correlations, but Ishida showed that this is incorrect when one takes into account full structure of hyperfine interaction tensor. Below, for example, is shown the expected response in the presence of certain assumed type of order on the Fe atoms (red).

Ishida said by changing angle of field, and noting that the experiment is sensitive to the fluctuations transverse to the field direction, they can verifiy that the order and low-frequency fluctuations are indeed of the stripe type shown in 1^st panel above.

Here are Ishida’s spin-lattice relaxation rate (T_1) data for various P dopings (Nakai et al., to appear in PRL). Ishida noted the large upturn in the (T_1T)^(-1) signal in the underdoped (magnetic) materials just above optimal doping. These fluctuations appear to grow until, near optimal doping, they reach a maximum (at this point T_N

And T_c are nearly equal— Ishida said whether they cross is not directly clear from NMR, and there is no neutron work because crystals are too small).

Data were fit to Curie-Weiss term b/(T+theta) plus constant offset. Variation of Curie-Weiss theta with doping shows theta goes through zero at/near optimal doping, where effective mass of one of the orbits from dHvA also appears to diverge. Ishida: this is consistent with claimed QCP at or near optimal doping. Alloul: there is no evidence for quantum cricitcal *behavior* from NMR alone; fact that theta-->0 at x=0.33 merely indicates T_N going to zero. Chubukov: is there low-T data showing QCP or only high-T extrapolated? Ishida: only the latter.

Ishida: qualitative behavior of Co-doped system is very similar, again with theta crossing zero around optimal doping. Some differences with K-doped material could be understood by changes in band structure, DOS at Fermi level which decreases as one e-dopes. Ishida reminded us that this emphasizes one great advantage of P-doping on the As site: unlike cuprates, or Co,K-doped 122 systems, changing “dopant” P does not change the DOS and one can examine changes in spin dynamics independently. Evidence: relative x independence of Knight shift.

Fernandes asked if there was evidence for re-entrant behavior of the superconductivity near the magnetic transition as in Co-doped systems. Ishida said he was not aware of any.

While the focus of this talk was not on SC state, Ishida pointed out that near optimal doping T-dependence was close to T^3 with low-T linear term (consistent with line nodes and some dirt). Benfatto pointed out that the overdoped data do not show the expected decrease of (T_1T)^(-1) below Tc at all, and Ishida said this was not understood.

Comparison with La-1111 (As NMR, powdered samples). Ishida pointed out that 1/T_1T changes by 2 orders of magnitude with F doping over a range x=0.05-0.15 where Tc changes by less than a factor of 2. Data show that while a similar, if weaker, increase in spin fluctuations occurs just above T_N in the magnetic phase, when one reaches the superconducting F doping concentrations (recall T_N has a first order drop to zero with doping in this system) these fluctuations are gone and (T_1T)^-1 is flat with T. His group is trying hard to understand these differences with 122 systems. Ishida claimed other 1111 systems, including those with higher Tc, show similar behavior when one subtracts rare earth magnetism (Ce,Sm, …).

Hirschfeld asked “devil’s advocate question” : if higher Tc superconducting family 1111 shows weaker spin fluctuations above the transition at optimal doping, is this not evidence that spin fluctuations are NOT responsible for superconductivity, or at least that another mechanism may be in play? Ishida agreed this was an important open question that they are investigating.