Beta-NMR of Isolated Lithium in Nearly Ferromagnetic Palladium

0.29 MB·English

by  T. J. Parolin

β-NMR of isolated Lithium in nearly ferromagnetic Palladium
T.J. Parolin,1 Z. Salman,2 J. Chakhalian,3,∗ Q. Song,4 K.H. Chow,5 M.D. Hossain,4
T.A. Keeler,4 R.F. Kiefl,2,4,6 S.R. Kreitzman,2 C.D.P. Levy,2 R.I. Miller,2
G.D. Morris,2 M.R. Pearson,2 H. Saadaoui,4 D. Wang,4 and W.A. MacFarlane1
1Department of Chemistry, University of British Columbia, Vancouver, BC, V6T 1Z1, Canada
7 2TRIUMF, 4004 Wesbrook Mall, Vancouver, BC, V6T 2A3, Canada
0 3Department of Physics, University of Arkansas, Fayetteville, AR, 72701, USA.
0 4Department of Physics and Astronomy, University of British Columbia, Vancouver, BC, V6T 1Z1, Canada
2
5Department of Physics, University of Alberta, Edmonton, AB, T6G 2G7, Canada

n 6Canadian Institute for Advanced Research
a (Dated: November23, 2006)
J
7 The temperature dependence of the frequency shift and spin-lattice relaxation rate of isolated,
1 nonmagnetic 8Li impurities implanted in a nearly ferromagnetic host (Pd) are measured by means
of β-detected nuclear magnetic resonance (β-nmr). The shift is negative, very large and increases
] monotonically with decreasing T in proportion to the bulk susceptibility of Pd for T > T∗ ≈ 100
i K. Below T∗, an additional shift occurs which we attribute to the response of Pd to the defect.
c
s Therelaxation rateis much slower than expected for thelarge shift and is linear with T below T∗,
- showing no sign of additional relaxation mechanisms associated with thedefect.
l
r
t PACSnumbers: 76.60.Cq,76.30.Lh,75.47.Np,73.21.Ac
m
.
t
a Elemental metallic[1] Palladium is on the verge of Pd foil, yielding a local measure of the magnetic charac-
m ferromagnetism as evidenced by its large temperature- teroftheLidefectthroughthenmrshiftandspinlattice
- dependent paramagnetic susceptibility χ. The chemical relaxation rate. Results in the film agree with measure-
d
and structural simplicity of Pd makes it a particularly mentsinthebulkfoilbutarebetterresolvedandhavean
n
o appealingexampleofanearlyferromagnetic(NF)metal. in situ reference fromthe Au layer,allowing anaccurate
c Efforts to understand NF metals have led to significant measureoftheshiftinPd. Wefindaverylarge,strongly
[ theoreticalprogress,suchastheadventofthespinfluctu- temperature-dependentKnightshiftK whichfollowsthe

∗
1 ationmodel(SFM)thathasrecentlybeenadaptedtothe host χ(T) down to T ≈ 100 K, with a deviation below
v nearly antiferromagnetic doped cuprates[2]. While the T∗ that we attribute to the response of Pd to the de-
4 SFMhas beensuccessfullyappliedto Pd,detailedcalcu- fect. In contrast, the spin-lattice relaxation rate T−1 is
1
9
lationsofquantitiessuchasχstillpresentachallenge[3]. remarkably slow and shows a simple Korringa(∝T) be-
3 ∗
Many recent attempts to understand the rich variety of haviour below T .
1
0 unconventional properties of nearly magnetic materials
nmr is a powerful technique; one of the few that can
7 are based on the paradigm of quantum criticality, where
0 proximity to a zero temperature quantum critical point reveal both the average magnetic behaviour and its mi-
/ croscopic inhomogeneity. Application of conventional
t (QCP)controlsthematerialpropertiesoverawiderange
a nmr to thin film heterostructures is limited by sensi-
of the phase diagram. In this context, Pd can be tuned
m tivity; however, β-nmr with a low energy beam of ra-
towardsanitinerantferromagneticgroundstateeitherby
- dioactive 8Li+ can be used in this case[8, 9]. In the nmr
d introducing dilute magnetic impurities[4] or by expand-
of metals, the relative frequency shift of the resonance,
n ing its lattice in an epitaxial heterostructure[5]. For ex-
o ample, refined studies of PdNi indicate the QCP occurs δ = (ν − νr)/νr, where νr is the reference frequency,
c is composed of two contributions δ = K + Korb: the
at a Ni concentration of only 2.5%[6], thus pure Pd is
v: in the realm of influence of this ferromagnetic QCP, al- Knight shift (K) resulting from Fermi contact coupling
Xi thoughitisclearlyaFermiliquid[1]. Thedefectresponse to the Pauli spin susceptibility of the conduction elec-
trons, and the temperature-independent orbital (chemi-
of a metal near a QCP is highly unconventional and
r cal) shift Korb. In conventional metals the spin relax-
a not yet understood. For example, it has been predicted
ation rate is linear with temperature, following the Ko-
that droplets of local order (interacting with the quan-
rringa Law, (T T)−1 ∝ K2. In contrast in NF metals,
tumcriticalenvironment)willbenucleatedbyapointlike 1
low frequency, long wavelength spin fluctuations modify
defect[7].
the Korringa Law, strongly enhancing the proportion-
In this Letter, we present an nmr study of an iso- ality constant at low temperature, and causing a high
lated atomic defect in pure Pd. In particular, we im- temperature deviation such that (T T)−1 is instead pro-
1
plant spin polarized radioactive 8Li and detect the nmr portionaltoχ[10],asseene.g.inTiBe [11]. Conventional
2
viatheparityviolatingweakβ-decay(β-nmr)inaAu(10 nmr of 105Pd[12, 13] showedthat K(T) follows the bulk
nm)/Pd(100 nm)/SrTiO heterostructure and in a thin χ(T), and the observedT−1 varies linearly below 150K.
3 1
2
Thepresentexperimentswerecarriedoutonthepolar-
ized low energy beamline at TRIUMF’s ISAC facility[8,
14]. The8Li+ probenuclei(I =2,τ =1.2s,8γ =6.3015
MHz/T) are implanted at an energy controlled by elec-
trostatic deceleration. The observed asymmetry of the
β-electron count rate (A) is proportional to the 8Li nu-
clear spin polarization. For resonancemeasurements, we
use a continuous beam of ∼ 106 ions/s focussed to a 3
mm diameter beamspot; measure the time-integrated A
for 1 s; then step the frequency of the small transverse
radiofrequency (RF) magnetic field H . The spin-lattice
1
relaxation rate T−1 is measured by pulsing the incident
1
ion beam (4 s pulsewidth) and monitoring the time de-
pendenceofAwiththeRFoff. A(t)isfittoanexponen-
tial form during and after each beam pulse[15]. All the
data were taken in a static external magnetic field H =
0
4.1 T[16]. The high sensitivity of the nuclear detection
allows us to work in the extremely dilute limit (the 8Li
concentrationis about1 in 1011), yielding the properties FIG. 1: β-nmr spectra of 8Li in the Au/Pd/STO film fit to
threeLorentzians. Theverticallineisthereferencefrequency
of isolated Li.
νr. The Pd signal is fit to two resonances corresponding to
Theresonancesweremeasuredina100nmfilmgrown
different stopping sites.
via e-beam deposition from a 99.99% source (Goodfel-
low)ontoanepitaxiallypolishedh100icrystallineSrTiO
3
(STO)substrateat60–80◦C.Thegrowthratewas0.5˚A/s
under a backgroundpressureof ∼10−8 Torr. It was sub- in isostructural Pd, but the two shifts are surprisingly
sequently capped in situ with 10 nm of Au. The Pd similar compared to Ag and Au, where they differ by a
factor of ∼ 2. In Ag and Au the more shifted O site
layer was found to be highly oriented in the h111i direc-
tion by x-ray diffraction. An implantation energy of 11 resonance is metastable, disappearing above ∼170 K as
keV was used as Monte Carlo simulations of the 8Li+ Li makes a thermally activated site change. The current
stopping profile(using TRIM.SP[17]) indicated thatthis data suggest a similar site change occurs in Pd above
would maximize the number of ions stopping in the Pd. 250 K. More measurements are required to confirm this
The T−1 data was collected in a 12.5 µm thick foil of site assignment. The two resonances are, however, very
1
99.95% Pd (Alfa Aesar). close, and they track one another as a function of T, so
we consider only the averageshift hereafter.
Three representative spectra are displayed in Fig. 1.
Resonances from 8Li in both Au and Pd were identi- A plot of δ (T) vs. χ(T) of bulk Pd[12] [inset Fig.
avg
∗
fiedbycomparisonwithpreviousmeasurementsinAu[18] 2(a)] shows a clear proportionality for T ≥ T . Ex-
andinPdfoil[19],aswellasthedependenceoftheirrela- trapolating this line to the zero of χ (or to the small
tive amplitudes on implantation energy. The shift of the van Vleck χ), we find Korb is quite small (40 ± 240
8LiresonanceinAuis+60ppmrelativetotheresonance ppm) as one would expect for a light atom. This es-
in the cubic insulator MgO[20]. We use this to infer the timate of Korb is rather uncertain due to the large ex-
reference frequency (in MgO) ν = 25.842 MHz (verti- trapolation, so rather than subtracting it from δ to ob-
r
cal line in Fig. 1) used throughout. Similar to previous tain K, we instead assume K ≈ δ. From the slope
measurements in unoriented polycrystalline Pd foil, the of this fit we extract the hyperfine coupling A us-
hf
shift is largeand negative, becoming more negative with ing K (T) = (A Z/N µ )χ(T); where N is Avo-
avg hf A B A
decreasingT,confirmingitis(a)intrinsictoLiinPdand gadro’s number, µ the Bohr magneton, and Z the co-
B
(b) predominantly isotropic. At 270 K, the Pd signal is ordination of the site assumed to be 6 (O-site), and ob-
clearlysplitintotworesonancesofnearequalamplitude. tain A = −1.50(2) kG/µ , a substantially smaller
hf B
The splitting is magnetic, rather than quadrupolar, im- magnitude than for Li in Ag[8] or Au[20]. For Pd, χ(T)
∗
plying two distinct Li sites of cubic symmetry. As T exhibits a characteristic maximum at T . Within the
is reduced, the less shifted resonance diminishes in am- SFM, this is due to thermal excitation of spin fluctua-
plitude as the more shifted line becomes predominant. tions producing effective magnetic moments. Once the
Previous β-nmr studies of 8Li in Ag[8] and Au[18] show temperature-inducedmoments reacha saturationampli-
∗
thatLitypicallyoccupiestwodistinctcubicsitesinFCC tude (∼ T ), they are then depolarized by higher en-
metals: theinterstitialoctahedral(O)andsubstitutional ergy thermal fluctuations, giving rise to the Curie-Weiss
(S) sites, each with a distinct shift. It is likely that the high T regime which is well described by Moriya’s self-
observed signals are due to 8Li stopping in similar sites consistent renormalization approach[21]. For Li in Pd,
3
FIG. 3: (color online). Normalized spin relaxation data and
fits. (inset) Temperature dependenceof (T1CT)−1.
ilar to the temperature independent relaxation found in
local moment paramagnets; however, a correct explana-
tionrequiresmodellingthespinfluctuationspectrum[13].
Thenearlytemperatureindependent(T TK )−1[Fig.
1C avg
2(c)] is consistent with NF behaviour and with the Pd
FIG. 2: (color online). (a) Temperature dependence of the T1[13], while the maximum in T1−1 near room tempera-
averageshiftof8LiinPd. (inset)Kavg vs. χ. K(T)obtained ture is not. Rather, it is likely due to a Li site change
from a fit to the linear region is represented by the solid red to a high temperature site characterized by a smaller
line in the main panel. (b) Temperature dependence of the shift and Korringa slope, as has been seen in other FCC
T−1 of 8Li+ in Pd foil. The red line is a fit to the low T
1 metals[20, 22].
data and the dashed blue line is the Korringa-predicted T−1
1
for Kavg= -885 ppm. (c) The product T1CTKavg is nearly We begin by discussing the originofthe largeshift K.
independent of T (squares; left scale). Some Kavg have been An impurity nucleus in a host metalis generallycoupled
interpolated/extrapolated from the data in (a). (T1T)−1 is to the conduction band via a transferred hyperfine cou-
also shown for comparison (circles; right scale). pling with its near neighbors. This interaction depends
on the overlap of the impurity atomic orbitals with the
neighboring conduction band orbitals, i.e., on the bond-
below T∗, the shift diverges from χ(T) of Pd (solid line ing of the impurity to the host. In Ag, Au and other
in Fig. 2(a)). simple metals this typically results in a small, positive,
We turn now to the spin-lattice relaxation. Previous temperature-independentK forLi. IncontrastinPd,we
results for 8Li in Ag demonstrated that the implanted observeashiftwhichisnegative,temperature-dependent
ionsarecoupledtotheconductionelectronsandobeythe and about 10 times larger. The first point is in fact ex-
KorringaLawquite closely[8]. The Pdfilm wasnotsuit- pected for a coupling arising from “s−d hybridization”
able for T measurements as a fraction of the Li always betweentheLi2sandPd4dorbitals[23],whilethelatter
1
stops in the Au overlayer. To avoid such contamination, two points result from the very large T dependent χ of
we instead measured T in a Pd foil which exhibits sim- Pd. A recent calculation predicts K for Li in Pd within
1
ilar resonances[19]. Representative relaxation data and a factor of 2 of the observed value[24]. The shift from
fits areshowninFig.3. The relaxationispredominantly such a coupling is isotropic, as observed.
∗
singleexponential,withasmall(<10%)backgroundsig- ThedeviationoftheLishiftK fromχbelowT reflects
nal from backscattered Li. The relaxation rates T−1 a local modification of χ due to the nonmagnetic defect.
1
are shown in Fig. 2(b). A linear fit below T∗ yields Incontrast,inPdnmr[12,13],theK−χscalingismain-
T−1= 1.22(2)×10−3s−1K−1T +4.5(9)×10−3s−1. To tainedtolowtemperature. IdenticalbehaviorforK ofLi
1
extract the purely Korringa relaxationT−1, we subtract in Pd foil[19] confirms the deviation is intrinsic. Indeed
1C
the small T independent term from the observed rate . such a response is similar to the evolution of χ in dilute
Above T∗, T−1(T) is clearly sublinear. This tendency alloys of PdAg[25], where 2.5% Ag is sufficient to elimi-
1
to saturation at high temperatures is qualitatively sim- nate the maximum in χ(T). In such alloys, despite their
4
dilution, it is difficult to rule out 4d band-filling as the
origin. In contrast, here the Li is in the dilute limit. We
note similar behavior appears in K(T) of the implanted
∗
Former address: Max-Plank-Institut fu¨r
µ+ inPd[26]. Thesesimilaritiessuggestacommonorigin
Festk¨orperforschung, D-70569, Stuttgart, Germany.
which is quite insensitive to the details of the nonmag-
[1] J.J. Vuillemin, N. Harrison, and R.G. Goodrich, Phys.
neticimpuritypotential. Suchanexplanationhasinfact Rev. B 59, 12177 (1999).
been suggested on theoretical grounds[27]. It would be [2] e.g. T.MoriyaandK.Ueda,Adv.Phys.49,555(2000);
interesting to test this theory with a detailedcalculation P.W. Anderson, Adv.Phys.46, 3 (1997).
of the Li+ defect in Pd. [3] B. Zellermann, A. Paintner, and J. Voitl¨ander, J. Phys.:
AmajordifficultyininterpretingPdnmrisdecompos- Condens. Matter 16, 919 (2004); P. Larson, I.I. Mazin,
ing K and T1−1 into orbital and spin contributions[13]. [4] Jan.CdrDan.Jg.leSainngdhW, P.Rhy.sS.cRotetv,.JB.A6p9p,l.0P64h4y2s9.3(620,0942)1.(1965);
For Li, we do not expect orbital effects since the atomic
A. Oswald, R. Zeller, and P.H. Dederichs, Phys. Rev.
electrons are in a state of zero orbital angular momen- Lett. 56, 1419 (1986).
tum. This is confirmed by the small size of Korb. Sim- [5] M.B. Brodsky and A.J. Freeman, Phys. Rev. Lett. 45,
ilarly, we expect the orbital contribution to T−1 to be 133 (1980); Z. Celinski et al., Phys. Rev. Lett. 65, 1156
negligible[28]. Inboth Ag andAu, the Lirelaxa1tionrate (1990); L.Chenget al.,Phys.Rev.B69,144403 (2004).
followstheKorringaLaw: (T T)−1 iswithinafactorof2 [6] Here the quantum critical point is the concentration x
ofvaluespredictedfromthesh1ifts. IfPdweresimilar,the Mwh.eNreictkhlaesfeertraolm.,aPghnyetsi.cRTeCv.iLsejtuts.t8s2u,p4p2r6e8ss(e1d99to9)0., see
Korringa T−1s should be much larger [e.g. dashed line
1 [7] A.J. Millis, D.K. Morr, and J. Schmalian, Phys. Rev.
Fig. 2(b)], and any additional orbital relaxation would Lett. 87, 167202 (2001).
only enhance this discrepancy. Millis et al. predict that [8] G.D. Morris et al.,Phys. Rev.Lett. 93, 157601 (2004).
a defect in an itinerant NF system will nucleate a local [9] T.A. Keeler et al., Physica B 374-5, 79 (2006), and in
order droplet with a magnetic fluctuation spectrum that preparation.
[10] T. Moriya and K. Ueda, Solid State Commun. 15, 169
is locally highly suppressed at low frequency[7]. While
(1974);M.T.B´eal-Monod,Phys.Rev.B28,1630(1983);
pure Pd may not be close enough to the QCP for this A. Ishigaki and T. Moriya, J. Phys. Soc. Jpn. 65 3402
theory to apply in detail, it is qualitatively consistent
(1996).
with ourobservationthat belowT∗,where the shift con- [11] H. Alloul and L. Mihaly, Phys. Rev. Lett. 48, 1420
tains a defect contribution, we obtain only very slow T- (1982).
linear relaxation. As the impurity contribution to the [12] J.A. Seitchik, A.C. Gossard, and V. Jaccarino, Phys.
total shift is relatively small (< 20%), the deviation in Rev. 136, A1119 (1964).
(T TK )−1 belowT∗isonlyapparentasasmallslope [13] M.TakigawaandH.Yasuoka,J.Phys.Soc.Jpn.51,787
1C avg (1982).
at low temperature. [14] Z. Salman et al.,Phys.Rev. B 70, 104404 (2004).
Inconclusion,the shiftandspin-latticerelaxationrate [15] Z. Salman et al.,Phys.Rev. Lett.96, 147601 (2006).
of highly dilute nonmagnetic 8Li impurities implanted [16] For thin films the demagnetizing factor ∼ 1, thus the
in NF Pd were measured via β-nmr. The shifts are internal field is equal to theapplied field.
isotropic, highly temperature-dependent, and scale with [17] W. Eckstein, Computer Simulation of Ion-Solid Interac-
the host χ above a characteristic T∗, below which they tions (Springer,Berlin, 1991).
[18] W.A. MacFarlane et al.,Physica B 326, 213 (2003).
continuetoincreaseuntilsaturatingatthelowestT. The
[19] T.J. Parolin et al.,Physica B 374-5, 419 (2006).
anomalousadditionalshiftatlowT isattributedtoade-
[20] T.J. Parolin et al.,in preparation.
fectresponseofthePd. TheKorringarelaxationismuch [21] T. Moriya, Spin Fluctuations in Itinerant Electron Mag-
slowerthanexpectedforthelargevalueoftheshift. The netism (Springer,Berlin, 1985).
product (T TK )−1 is approximately T-independent [22] e.g. Z. Salman et al.,submitted to Phys. Rev.B.
1C avg
as expected for a NF metal. It will be interesting to fol- [23] H.Akaietal.,Prog.Theor.Phys.Suppl.101,11(1990);
lowtheevolutionofthesepropertiesinboththinnerfilms C.D.Gelatt,Jr.,A.R.Williams,andV.L.Moruzzi,Phys.
Rev. B 27, 2005 (1983).
(where finite size effects are expected to play an impor-
[24] M. Mihara et al.,HyperfineInteract. 158, 361 (2004).
tant role) and in alloy films and heterostructures closer [25] R. Doclo, S. Foner, and A. Narath, J. Appl. Phys. 40,
to quantum criticality.
1206 (1969).
We acknowledge the technical assistance of R. [26] F.N.Gygaxetal.,SolidStateCommun.38,1245(1981).
Abasalti, B. Hitti, D. Arseneau, S. Daviel, H.-U. Haber- [27] D.R. Grempel, Phys.Rev.B 27, 4281 (1983).
meier and M. Xu, and funding from NSERC Canada. [28] P. Lee and N. Nagaosa, Phys.Rev.B 43, 1223 (1991).
TRIUMF is funded in part by the NRC.