Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Rspa 1951 0062

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

T he th eo ry o f th e n u clear hyperfine stru c tu re of

p aram ag n etic resonance sp ectra in th e copper


T u tto n salts
B y A . A b r a g a m a n d M. H. L. P r y c e
Clarendon Laboratory, University of Oxford

(Communicated by B. Bleaney, F.B.S.— Received 11 September 1950)


Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

The general theory of a previous paper is applied to the copper Tutton salts. It is found that
unless account is taken of the effect of admixed configurations containing unpaired ^-electrons,
agreement with experiment is impossible. Allowance for unpaired ^-electron configurations
gives good agreement, both for anisotropy and magnitude of the effect. The effect of the
nuclear quadrupole moment, though small, is important, and the paramagnetic resonance
method is in principle perhaps capable of giving a more accurate value of this nuclear constant
than optical methods.

1. I n t r o d u c t io n

In a previous paper (Abragam & Pryce 1951, to be referred to as (A)) a general theory
of the nuclear hyperfine structure of paramagnetic resonance spectra has been given.
In the present paper this theory is applied to the copper Tutton salts, i.e. the double
sulphates CuS04ilf2S04. 6H20, where M is a monovalent ion. A preliminary account
of some of the results has already been given (Abragam & Pryce 1949).
According to crystallographic data, the water octahedron surrounding the copper
ion Cu++ is distorted along an axis passing through two opposite vertices, leading to
the assumption of a crystalline field V of tetragonal symmetry.
From spectroscopic data it follows th at the lowest level of the free ion Cu++ is
2D, and in first approximation the wave function of the ion in the solid state belongs
to this manifold (approximation A of (A)).
It has been shown in (A) that under these circumstances (tetragonal symmetry,
total spin the energy levels of the copper ion in a magnetic field are the eigenvalues
of the ‘spin’ Hamiltonian:
W = /}gnHl SK+figL(HxS9 + HyS1l) + A S .Iz
+ B ( S J X+ Sv I,) + Q(I\ - * /(/+ 1)) - y/?„(H. I).
The measurements of Ingram (1949) show that the observed pattern can indeed be
accounted for by using a Hamiltonian of the form (1*1), with the following typical
values for the different constants (for A and, I?, the magnitudes are known, but not
the signs):
011-2*4, g±~ 2-1,
| A|~ 1 3 x 10-3 cm.-1,
3-5 x 10-3cm.-1, in the potassium Tutton salt,
2 x 10-3 cm.-1, in the ammonium Tutton salt,
x 10-3 cm.-1, in the rubidium Tutton salt.
[ 164 ]
The theory of the nuclear hyperfine structure 165
There is no precise information concerning the constant Q. A preliminary reduction
of the data indicates | Qj= 0*9 x 10-3 cm.-1. The last term in W is negligible
magnetic fields hitherto available.
Copper has two isotopes, both with nuclear spin = §, and nearly equal magnetic
moments. The constants therefore represent an average over unresolved, or only
partly resolved, lines.
The tetragonal symmetry is only approximate:
gx ~2-05, 2-10.
However, this departure from tetragonal symmetry does not alter the main features
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

of the spectrum and can be overlooked in first approximation.


Our main problem is to calculate the different constants in the Hamiltonian
W which are given by the formulae (3*5) to (3-9) of (A), in order to compare them with
the experimental values. As already pointed out in (A), it is not possible to calculate
completely these constants from first principles, partly because of our very imperfect
knowledge of the wave function of the free ion, knowledge required for the calculation
of the constants k, A and r-3, and chiefly because of our ignorance of the crystalline
potential V. The best we can aim at is to calculate these constants by using results
taken from other experiments. The knowledge of the fine and hyperfine structure
of atomic spectra enables us to calculate A and r-3 and gives indications about k . The
energy levels of the ion in the crystalline potential V could in principle be obtained
by infra-red spectroscopy of the crystal, but in this domain experimental evidence
is scanty. Attempts such as Polder’s to calculate V a priori are unconvincing, and,
a t present, crystalline symmetry considerations and magnetic data are the main
guides for the choice of V.
2. A p p r o x i m a t i o n B
I t has already been stated in (A) th at stationary states of the free ion, or eigenstates
of WF, can be represented with reasonable accuracy by a single configuration.
Departures from this approximation (approximation of can be evaluated by
using perturbation theory. We shall adopt this approximation as a starting point,
and take for the lowest level of the free ion Cu++ the term 3 with one electron
missing from the complete 3 dshell.
A calculation of the g values, resting on these assumptions and treating the 3
electrons as a positive hole in a complete shell, has been given by Polder (1942), and
can be summarized as follows: A cubic field splits the term into a lower lying
orbital doublet, spanned by the orbital (^-functions (x2—y 2) and (3z2—r2), and an
upper triplet spanned by xy,yz, zx. A tetragonal field splits the d
a lowest level without orbital degeneracy. If we make the assumption, substantiated
by crystallographic evidence, th at the water-octahedron is elongated rather than
flattened along the tetragonal axis, we must take x2—y2 as the wave function of the
lowest level. Figure 1 shows how the 2D level is split by the crystalline field. The
field being mainly cubic, the tetragonal splittings — and Ex —E0 are likely to
be much smaller than the cubic splitting A According to Polder they are of the
same order of magnitude, but this assumption is not supported by any experimental
166 A. Abragam and M. H. L. Pryce
evidence. In order to get the gr’s we must calculate the principal values of the tensor
A given by (3*5) of (A). It is easily found that
1
L33 '= et jji > Au —A22 ~ E 9
Jha
'2 — "JlJn
0

8A 2A
and thence 011 = 2 Ez - E 0 * 0j. = 2 E9—E0 ( 2 - 1)

A more accurate calculation gives


a 8A 4A2 8A2 A2
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

to - Et - E , + (Et - E , ) ‘ ( E ,- E „ ) ( E ,- E <>)+ (EZ- E 0f ,


( 2 *2 )
2A 2A2 A2
H ~ E ,- E „ \Et - E „ f (Et - E ') ( E ,- E » Y )

free ion + cubic field+ tetragonal field+ L S coupling


F igure 1. Energy levels of Cu++ in the Tutton salts.

If we take E3—E0 = E3—E0 = A E ,we get the approximat


A can be calculated by Goudsmit’s theory or, better, deduced from the atomic-energy
levels of the free ion Cu++, now available (Shenstone, unpublished). One finds:
—A = 828 cm.-1. If we take gy= 2*4, gL = 2*1, we find
E3—E q~ E 2—E 0= 17,000 cm.-1.
These energy differences lead to values for the cubic field which agree reasonably well
with those chosen for other salts of the iron group, but provide little information
concerning the magnitude of the tetragonal field. Optical absorption in solutions of
cupric salts shows a broad peak at 12,300 cm.-1, but no other absorption resonance
below 32,000 cm.-1. As the colour of the crystals is much the same as of the solution,
we take this to imply that A Ei
s actually about 12,000 cm.-1. T
crepancy, which may perhaps indicate that A is smaller in the crystal than in the
free ion (cf. (A)).
It is easy, though not very enlightening, to account formally for the inequality
between the values gx and gy by assuming for the field a small rhombic component
The theory of the nuclear hyperfin structur
with the same axes as the cubic field. The wave function of the lowest level takes the
form: ^(3) cos a (a:2— y 2)+sina(3z2—r2), where a is a small constant, and the level
Ez gets split into two sublevels E zand E H
Z. The follo
established
2A
9* = 2 ~ - j - r (cos a + V(3) sin a)2,'

2A
9V = 2 - ^ 7 ^ ( c o s a - V ( 3 ) s i n a ) 2,^ (2-3)

8A
2 cos2a.
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

9z -
^2~E 0

It is easy to choose for a and E z —E z small values which will fit the experimental
values of the g’s.
On the whole, Polder’s theory of the (/-values, possibly supplemented by the
introduction of a small rhombic field, gives very reasonable agreement with experi­
ment. We shall see presently th at for the hyperfine structure the situation is quite
different.
The constants A and B are obtained from the formulae (3*7) a n d '(3*8) o
these formulae k = 0 since in our approximation B, the ionic wave fun
represented by the configuration 3d9, which does not contain unpaired s-electrons.
We get
-4 = —2 —3£Z33+ 3£Aw33) P ,
(2*4)
B= {g±—2 —3£
Using the definitions (2*9), (3*1) and (3*6) of (A) for l and u, and £ = gj, we find

A ( ?l1 2 * 7(£a-iy)p’j (2*5)

Taking (2*1) into account we get

^ = (S7ll- 2 + l( ?J. - 2 ) - f ) P ~ - 0 - 1 3 P , 1
B = ( g L - 2 - A(sr± - 2) + f ) ~ + 0-36sP. / (2'6)
A first check of the theory is to compare the experimental and calculated values of
the ratio j A /B |.
For | /Bcalc
A | we find 0-36, while the experimental values lie between 4 and 10.
The discrepancy between theory and experiment is extreme, and indicates th at we
must look for an improvement on our approximations.
We now want to estimate the constant P. The value of y is known with extreme
precision from nuclear resonance experiments (Pound 1948), and we shall take for
it the value of 1*53, being an average for the two isotopes. To evaluate 3 we can
tentatively use Hartree wave functions (Hartree 1936), which are available for the
monovalent copper ion Cu+ 3d10; graphical integration gives r~z =7*25 atomic units,
168 A. Abragam and M. H. L. Pryce
whence P = 0*035 c m r 1. Another evaluation of P is provided by the optical hyperfin
structure of the free atom Cu. Ritschl (1932) has measured the hyperfine structure
intervals of Cu for the following levels:
3dHs22D* of = 0*022 cm.-1,
3d3*94s22Dia} = 0-060 cm.-1.
The two 4s electrons in 3d94s2give no contribution to the hyperfine
is due to the missing electron in the 3d shell. Using the well-known formula
aj = Pl{l +1 )lj(j + 1) we get:
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

From oq P = 0*032 cm.-1, from af P = 0*0375 cm.-1.


We take for P the average value 0*036 cm.-1.* The fact th at calculations' of P using
the wave functions of Cu+ 3d10 or the optical hyperfine structure of Cu 3d94s2lead to
approximately the same result gives us confidence in extrapolating this result to
Cu++3d9.
We get for the constants A and B:
A = —0*0046 cm.-1,
B = 0*0131 cm.-1,
in complete disagreement with experiment.

3. I nfluence of s-electrons
We are thus led to the conclusion that the approximation B is not sufficiently
accurate and that beside 3d9 the correct wave function of the ion contains admixtures
from higher configurations (Abragam 1950).
I t is a well-known fact in optical spectroscopy th at configurations with unpaired
s-electrons (we shall for brevity call them ^-configurations) generally have a much
wider hyperfine structure than any other configurations, and this disparity is con­
siderably enhanced in the solid state, where the quenching of the orbital momentum
reduces by several times the hyperfine structure of all but s-configurations.
The nearest configuration 3d94s being more than 70,000
admixture in the wave function of configurations other than 3d9 milst be small, and
it is reasonable to neglect all but s-configurations. While giving up the approximation
B, we still retain the approximation ,i.e. the assumpt
A
of the ion in the solid state is an eigenfunction corresponding to the lowest level
2D of the free ion, and the only admixtures from higher configurations will also be
2D terms of same parity as 3d9, i.e. even. Little can be gained by going beyond this
approximation and taking into account the coupling of excited configurations by the
crystalline field. I t is easy to show th at no ^-configuration can be coupled to 3d9 by
the cubic field, and the effect of the tetragonal field is negligible in comparison with
the configuration coupling included in the approximation A.
* I f w e take into account a /c-term, as in the n ext section, we find P = 0*036 cm .-1, k —0*36,
for Cu3d94s2.
The theory of the nuclear hyperjine structure 169
The wave function of the ion has the form
2D(3d9) + a12D1+...+oci 2Di + ..., ( 3 - 1)
where 2Di is a term originating from an ^-configuration and cci a small coefficient (we
neglect the normalization factor). A calculation by Abragam & Horowitz (to be
published) shows th a t the im portant terms are ( )3 2D, in which
an electron from the normally occupied ms state is raised to the unoccupied ns
state. They give contributions to the unpaired spin density at the nucleus which
are linear in the a ’s, while other terms, such as SdHs 2D or m s.n s^S ) 3 2D give
quadratic contributions, which are presumably much smaller.
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

Formally we account for the presence of unpaired 5-electrons by giving a non-zero


value to the constant k . The following approximate formula results:

~ ^3j [a 3s4s(®3s®4s)i + ...], (3'2)

where a^, are the optical hyperfine structure interval factors of the 3s and
4s electrons.
The new values for A and Ba
re
A = ( —x —0-13)P,
B = ( —x + 0-365) P .
It is obvious th a t a negative value of k makes the disagreement with experiment
still worse. On the other hand, by taking for instance k = 0-25, we find
A calc. = ~ ° '013 cm .-1, P calc = + 0-004 cm."1,
which is in reasonable agreement with experiment.
I t is interesting to notice th at this agreement is reached simultaneously for A and
B by introducing a single constant k . The data we have chosen, however, are
a rough average for a group of salts. A more decisive test is furnished by taking
the data for copper ammonium sulphate, which are known experimentally with
reasonable accuracy (Bleaney, Penrose & Plumpton 1949; Ingram 1949). The
three principal ^-values are 2-45, 2-12 and 2-06. It will be seen th at the assumption
of tetragonal symmetry is only poorly satisfied, but for the purpose of comparison
we adopt the following values:
0,1 = 2-45, g± = 2-09;
A = -0-0134, B = 0-0020 cm .-1.
The value of B is uncertain because of the departure from tetragonal symmetry,
and because of the quadrupole effect discussed in §4. W ith these values we find
that a value of 0-28 for k gives
^caic. = 0-013l5 £ calc. = 0-0027 cm."1.
The last digit in these values is not significant, however, for an error of g by 0-01
will cause an error of about 4 in the last place. The fit is therefore about as good
as the present experimental accuracy.
170 A. Abragam and M. H. L. Pryce
Further confirmation comes from the consideration of the copper fluosilicate
CuSiF6. 6H20. I t has been shown (Abragam & Pryce 1950) th at in approximation B
the calculated values of A and B for the fluosilicate are
while the experimental values are | A | ~ 0*0021, 0*0028 cm.-1. Since k is
essentially a feature of the free ion Cu++ there must be little difference between its
values in the Tutton salts and in the fluosilicate. A slightly different value = 0*30
gh es A~ ~
B —0*06P ~ —0*0021 cm.-1 for the fluosilicate.
Here again a reasonable agreement is reached. The new values of A and B are
extremely sensitive to the exact values of k and of the s and should not be taken too
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

literally. (In the fluosilicate, varying gby 1 % leads


Still it seems rather remarkable th at by applying nearly the same correction (as
required by the theory) in two salts where the circumstances are so utterly different,
one gets in both the right magnitude and the right anisotropy for the hyperfine
structure.
It appears reasonable to assume th at the main contribution to (3*2) comes from
the 3s4s.
. 3d9 configuration. Using the value k = 0*3, and the values — 4 cm.-1
(calculated from Hartree wavefunetions for Cu+), ais = 0*2 cm.-1 (from the hyper­
fine structure of the 3d10 4=s2S term of Cu, Tolansky 1938), we find t
a 3s4s °f the admixture is about 0*006, corresponding to a partial probability of
4 x 10-5. I t is seen that a very minute admixture is capable of having a profound
effect on the hyperfine structure.

4. T he quadrupole effect
The factor Q of formula (1 •1) is given by

Q = 41(27 - 1)^ = - T ie*q h <A><2' 10>’ <2' U >’ <3'3>’ <3'9)


(tj = —2j, / =§). It is convenient to measure Q and q' in 10-3em.-1, 1/r3 in atomic
units, q in 10-24cm.2. In these new units:

« = - 113^ -

Since 5-electrons give no contribution to the quadrupole effect we can use approxima­
tion B which gives l33 = 2
e = - 1-13^.
If we take | Q \ =0*9 and r-3 = 7*2 calculated earlier we find
|? | = 0*11,
in fair agreement with the optical value (q= —0*2
isotopes (Brix 1949)). The paramagnetic evidence strongly suggests that both
isotopes have nearly equal quadrupole moments, however, in contradiction to Brix s
conclusion.
The theory of the nuclear structure 171

5. Conclusion
I t appears th at the main features of the hyperfine structure in the copper salts
are explained a t least qualitatively. Unfortunately, it does not seem possible to
estimate with any accuracy the admixture coefficient without very heavy numerical
calculations. On the experimental side a determination of the signs of and B would
provide the theory with a further check.
This would perhaps become possible if much higher fields and frequencies were
available, by measuring an interference effect between the terms containing A and
B and the hitherto negligible term — y/3n
H . I.
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

A ppendix
We fist below several earlier and unsuccessful attem pts which we made to explain
the observed hyperfine structure in the Tutton salts.

Rhombic field
The idea th a t a strong rhombic component in the crystalline field could explain
the facts, stemmed from the following remark: If the lowest state were not x2—y 2
but 3z2—r2 (which corresponds to a flattening of the water octahedron along Oz), the
signs of ln and Z33 would be reversed, and j A | would become bigger than j | in
conformity with experiment. This, however, would upset the values, leading to the
unacceptable value = 2. Since in a rhombic field the lowest state is a superpositibn
of x2—y2 and 3 z2—r2, we had hoped th a t a suitable choice of the relative w
of these two functions would lead to a simultaneous approximate agreement for the
g’s and the hyperfine structure. But a detailed investigation showed th a t it was
impossible.
Motion of the water molecules
If the water octahedron, instead of being permanently elongated as in Polder’s
theory, had two equilibrium positions, one flattened and the other elongated, with
continual flopping between them (as in the ammonia molecule), the observed
physical quantities would be weighted averages of values corresponding to these
two equilibrium positions. Since we suppose expectation values themselves rather
than the wave functions cross-terms disappear and the result is not the same as in
the first case. The necessity of assuming motion of the water molecules in the
fluosilicate added weight to this hypothesis. Here again the agreement with
experiment proved impossible.

Covalent bonding hypothesis


Here the forces between the copper ion and the water molecules were supposed
to be covalent rather than electrostatic. Two of the six water molecules, farther
removed from the ion than the others, are assumed to be beyond the range of overlap
covalent forces. A strong square bond is assumed between the central ion and the
oxygens of the four remaining molecules. According to Pauling (1940), the strongest
172 A. Abragam and M. H. L. Pryce
square bond is constructed in electron pairing method by hybridizing the four
electronic orbitals s, p x,p y, dxy.
I t is easy to show th a t under those assumptions there is only one unpaired electron
responsible for the magnetic properties of the salt, and th at it is in a state pz. I t can
also be shown th at Mulliken’s method of molecular orbitals leads to the same result.
In the state pz the constants lzz and Zu are respectively £ and —§ and $ satisfactory
anisotropy can be obtained for the hyperfine structure.
A closer investigation shows, however, that this structure also leads to = 2.
Finally, this covalent structure has different symmetry properties from Polder’s
Downloaded from https://royalsocietypublishing.org/ on 27 February 2022

ionic structure, and no resonance can exist between them.


In spite of its many attractive features, the covalent hypothesis had therefore to
be rejected.
R eferences
Abragam, A. 1950 Phys. Rev. 79, 534.
Abragam, A. & Pryce, M. H. L. 1949 Nature, 163, 992.
Abragam, A. & Pryce, M. H. L. 1950 Proc. Phys. Soc. A, 63, 409.
Abragam, A. & Pryce, M. H. L. 1951 Proc. Roy. Soc. A, 205, 135.
Bleaney, B., Penrose, R. P. & Plumpton, B. I. 1949 Proc. Roy. Soc. A, 198, 406.
Brix, P. 1949 Z. Phys. 126, 725.
Hartree, D. R. 1936 Proc. Roy. Soc. A, 157, 490.
Ingram, D. J. E. 1949 Proc. Phys. Soc. A, 62, 664.
Kopfermann, H. 1940 Kemmomente. Leipzig: Akademische Verlagsgesellschaft.
Pauling, L. 1940 The nature of the chemical bond. Ithaca, N .Y .: Cornell University Press.
Polder, D. 1942 Physica, ’s Orav., 9, 709.
Pound, R. Y. 1948 Phys. Rev. T&, 523.
Ritschl, R. 1932 Z. Phys. 79, 1.
Tolansky, S. 1938 Proc. Phys. Soc. A, 50, 831.

You might also like