Earth Planets Space, 51, 10191022, 1999

A correction to Bahrs phase deviation method for tensor decomposition

Erno Pracser1 and Laszlo Szarka2

1 Lorand Eotvos Geophysical Institute, H-1145 Budapest, Kolumbusz u. 17-23, Hungary

2 Geodetic and Geophysical Research Institute, H-9401 Sopron, POB 5, Hungary

(Received November 9, 1998; Revised February 17, 1999; Accepted April 12, 1999)

For models having moderate departures from the basic distortion model (from the so called principal superim-
position model, that is where local three-dimensional and regional two-dimensional structures are superimposed)
a special tensor decomposition method (the so called phase deviation method) was suggested by Bahr (1991). As
far as we know, this technique has never got a wide field application in the field. In a careful examination of the
suggested solution, an error was observed in the original derivation of the formulas. In this paper Bahrs (1991)
solution is corrected. Using the new equations, more understandable and interpretable results are obtained, as it is
illustrated on synthetic examples.

 
1. Introduction a11 a12
For the interpretation of a measured impedance tensor where a = is a real, frequency-independent ma-
a21 a22
according to Bahr (1988)one must ask, whether all ele- trix, representing the distortion due to small, localized, near-
ments of the measured tensor have the same phase. If they surface anomaly. Z T E and Z T M are the principal impedances
do, the regional conductivity is purely depth dependent and it for the 2D regional structure. According to the condition
is sufficient to split the impedance tensor into a real distortion set up by Bahr (1991), the phases between the elements in
matrix and a scalar normal impedance. both columns of the tensordue to the effect of the phase-
If only the two elements in each column of the measured sensitive regional skew (Bahr, 1991)differ by the same
tensor have equal phase values, the regional conductivity phase deviation angle .
is two-dimensional. The tensor decomposition solution for The strike angle is found from the two conditions for the
this so called principal superimposition model is found in the two columns of the impedance tensor of Eq. (1), whereas the
paper by Bahr (1988). two variables and are to be resolved. From a comparison
If the regional conductivity distribution is not perfectly of the real and imaginary parts of elements in both columns,
two-dimensional, a phase difference will appear between two equation are obtained:
the two elements of the same column. For this problem
assuming a moderate departure from the principal superim- Re Z x  x  cos + Im Z x  x  sin
position modela solution was given by Bahr (1991), which Re Z y  x 
he calls the phase deviation method. We found an error Re Z x  x  sin + Im Z x  x  cos
= , (2a)
in the original derivation of the formulas. In this paper we Im Z y  x 
first give a brief description of the phase deviation method, Re Z y  y  cos Im Z y  y  sin
then we present the corrected solution. Finally a comparison Re Z x  y 
between the original and the corrected solution is given.
Re Z y  y  sin + Im Z y  y  cos
= . (2b)
Im Z x  y 
2. Brief Description of Bahrs (1991) Phase Devi-
ation Method In another form:
In case of moderate departures from the principal superim- cos (Re Z x  x  Im Z y  x  Re Z y  x  Im Z x  x  )
position model, Bahr (1991) represented the measured mag- + sin (Re Z x  x  Re Z y  x  + Im Z x  x  Im Z y  x  ) = 0
netotelluric tensor Z in the coordinate system of the regional (3a)
2D structure as and
 
a12 Z T M ei a11 Z T E cos (Re Z y  y  Im Z x  y  Re Z x  y  Im Z y  y  )
Z= , (1) sin (Re Z x  y  Re Z y  y  + Im Z x  y  Im Z y  y  ) = 0.
a22 Z T M a21 Z T E ei
(3b)
To solve these equations, Bahr (1991) introduced
Copy right
c The Society of Geomagnetism and Earth, Planetary and Space Sciences
(1) two commutators between the complex numbers C1 and
(SGEPSS); The Seismological Society of Japan; The Volcanological Society of Japan; C2 as follows:
The Geodetic Society of Japan; The Japanese Society for Planetary Sciences.
[C1 , C2 ] = Im(C2 C1 ) = Re C1 Im C2 Re C2 Im C1 , (4a)
1019
1020 E. PRACSER AND L. SZARKA: A CORRECTION TO BAHRS DECOMPOSITION

{C1 , C2 } = Re(C2 C1 ) = Re C1 Re C2 + Im C1 Im C2 . 3. Problem

(4b) We have found that in Eqs. (9a) and (9b) a further term
In this way Eqs. (3a) and (3b) have the following simpler must exist. In addition, the indices in Eq. (10d) are mixed
form: up and some other misprints should be corrected, too.
We think, the origin of the problem in the derivation by
cos [Z x  x  , Z y  x  ] + sin {Z x  x  , Z y  x  } = 0, (5a)
Bahr (1991) must be an erroneous transformation of the com-
cos [Z y  y  , Z x  y  ] sin {Z y  y  , Z x  y  } = 0, (5b) mutator {C1 , C2 }.
and (2) he defined four so called modified impedances
4. Correction
Z x x + Z yy
S1 = , (6a) Since
2
Z x y + Z yx C2 C1 = a2 ei2 a1 ei1
S2 = , (6b)
2 = a1 a2 cos(2 1 ) + ia1 a2 sin(2 1 ), (12)
Z x x Z yy
D1 = , (6c)
2 the commutative laws for the two commutators are:
Z x y Z yx
D2 = . (6d)
2 {C1 , C2 } = {C2 , C1 }, (13a)
Their transformation into a new coordinate system which [C1 , C2 ] = [C2 , C1 ]. (13b)
is rotated by an angle yields
It is inevitable to express {Z x  x  , Z y  x  } in details:
Z x  x  + Z y y
S1
= = S1 , (7a)
2 {Z x  x  , Z y  x  } = {S1 + D1 cos 2 + S2 sin 2,
Z x  y + Z y x 
S2 = = S2 cos 2 D1 sin 2, (7b) D2 D1 sin 2 + S2 cos 2}
2
Z x  x  Z y y = {S1 , D2 } {D1 , D2 } cos 2
D1 = = D1 cos 2 + S2 sin 2, (7c) {S2 , D2 } sin 2 {S1 , D1 } sin 2
2
Z x  y Z y x  {D1 , D1 } sin 2 cos 2
D2 = = D2 . (7d)
2 {S2 , D1 } sin2 2 + {S1 , S2 } cos 2
Therefore Z x  x  , Z x  y  , Z y  x  and Z y  y  are a function of : +{D1 , S2 } cos2 2
Z x  x  = S1 + D1 = S1 + D1 cos 2 + S2 sin 2, (8a) +{S2 , S2 } sin 2 cos 2 (14)
Z x  y  = S2 + D2 = D2 + S2 cos 2 D1 sin 2,
(8b) = ({S1 , D1 } + {S2 , D2 }) sin 2
Z y  x  = S2 D2 = D2 + S2 cos 2 D1 sin 2, (8c) +({S1 , S2 } {D1 , D2 }) cos 2
Z y  y  = S1 D1 = S1 D1 cos 2 S2 sin 2. (8d) +({S2 , S2 } {D1 , D1 }) cos 2 sin 2
By using the commutators (4a and b) and the modified {S1 , D2 } + {D1 , S2 } cos2 2
impedances (8a, b, c and d) Bahr (1991) expressed Eqs. (5a) {S2 , D1 } sin2 2.
and (5b) as:
Sinceaccording to (13a){S2 , D1 } = {D1 , S2 }, the sum
A sin 2 + B cos 2 + C + E cos 2 sin 2 = 0, (9a) of the last three terms in Eq. (14) is
A+ sin 2 + B + cos 2 + C + + E + cos 2 sin 2 = 0,
(9b) {S1 , D2 } + {D1 , S2 } cos2 2 {S2 , D1 } sin2 2
= {D1 , S2 } {S1 , D2 } 2{S1 , D2 } sin2 2 (15)
where for the first equation (if indices 1 refer to the terms
with cos and indices 2 refer to those with sin ) and not {D1 , S2 } {S1 , D2 } as in Bahr (1991).
A = A1 + A2 = ([S1 , D1 ] + [S2 , D2 ]) cos According to Eq. (15) and to the corresponding relation-
+({S1 , D1 } + {S2 , D2 }) sin , (10a) ship between the two elements in the right column, Eqs. (5a)
and (5b) with two unknowns: and , have the following
B = B1 + B2 = ([S1 , S2 ] [D1 , D2 ]) cos form:
+({S1 , S2 } {D1 , D2 }) sin , (10b)
C = C1 + C2 = ([D1 , S2 ] [S1 , D2 ]) cos (A1 + A2 ) sin 2 + (B1 + B2 ) cos 2 + (C1 + C2 )
+({D1 , S2 } {S1 , D2 }) sin , (10c) +E 2m cos 2 sin 2 F2m sin2 2 = 0, (16a)
E = E 2 = ({S1 , S1 } {D2 , D2 }) sin , (10d) (A1 A2 ) sin 2 + (B1 B2 ) cos 2 (C1 C2 )
+E 2m cos 2 sin 2 F2m sin2 2 = 0, (16b)
and for the second equation
A+ = A1 A2 , (11a) where A1 , A2 , B1 , B2 , C1 and C2 are the same as in Eqs.
(10a), (10b) and (10c), and E 2m and F2m are as follows:
B + = B1 B2 , (11b)
C + = C1 + C2 , (11c) E 2m = ({S2 , S2 } {D1 , D1 }) sin , (17a)
+
E = E 2 = E. (11d) F2m = 2{D1 , S2 } sin , (17b)
 E. PRACSER AND L. SZARKA: A CORRECTION TO BAHRS DECOMPOSITION 1021

where the superscript m indicates modifications to the orig-

inal equations.
In the following, all steps of the solution are the same
as those of Bahr (1991), though the results are somewhat
different. From the sum and difference of Eqs. (16a) and
(16b), and can be derived from the following equations:

(a1 sin 2 + b1 cos 2) cos

+(c2 + e2m cos 2 sin 2 f 2m sin2 2) sin = 0, (18a)
c1 cos + (a2 sin 2 + b2 cos 2) sin = 0, (18b)

where the small case letters refer to the corresponding quan-

tities, without cos and sin .
Finally

1 b1 a2 + a1 b2 + c1 e2m
tan(21,2 ) =
 2 a1 a2 c1 c2 + c1 f 2m
1 (b1 a2 + a1 b2 + c1 e2m )2 b1 b2 c1 c2
m 2
4 (a1 a2 c1 c2 + c1 f 2 ) a1 a2 c1 c2 + c1 f 2m
(19)
Then can be determined from Eq. (18a) or (18b).
In the original solution (equation (30), Bahr, 1991), the
terms with f 2m are missing. Furthermore, e2 in the original
solution is not the same as the modified e2m .

5. Mathematical Discussion
According to Bahr (1991), this phase deviation method is
valid in cases where the phase sensitive regional skew and
the regional one-dimensional indicator do not vanish.
Their original definitions are as follows (Bahr, 1988,
1991):
1 1
(|[D1 , S2 ] [S1 , D2 ]|) 2 C2
= = , (20a)
|D2 | |D2 |
1
(|[D1 , S2 ]| + |[S1 , D2 ]|) 2
= . (20b)
|D2 |

If = 0, the C1 = c1 = 0. In Eqs. (18a) and (18b) if

c1 = 0, there are two mathematical cases:
1) sin = 0, which leads directly to

b1
tan 2 = .
a1

2) sin = 0. In this case

b2
tan 2 = .
a2

Since = 0 means a perfect two-dimensional regional

structure, this second case does not have any physical mean-
ing. Consequently in Eq. (19) it is a reasonable preference
to select from 1 and 2 the root associated with minimal
solution, as it was suggested by Bahr (1991).
The above analysis tells that Eq. (19) does remain if = 0.
The only case when Eq. (19) does not work, is if the regional
one-dimensional indicator is zero. With = 0 no strike
angle is obtained and only one single impedance value can
be recovered from the measured tensor.
1022 E. PRACSER AND L. SZARKA: A CORRECTION TO BAHRS DECOMPOSITION

6. Comparison with the Original Solution decomposition formulas.

Equation (19) in this paper and equation (30) in the paper
by Bahr (1991) may result in quite significant differences in
the regional strike estimates, as it is illustrated in Table 1.
In our numerical example an impedance tensor having rea-
sonable and values and some different distortion tensors
were selected. The tensor was rotated from its principal di-
rection with an angle of = 30 and the original strike
direction had to be reconstructed, by using the two formulas.
As it is shown in Table 1, while Eq. (19) found the correct
value ( = 30.0 and = 6.0 ), by using the original
formula = 38.3 and = 3.9 were obtained. References
When the same tensor was rotated with = +30 , Eq.
(19) gave again the correct value ( = 30.0 as expected),
the discriminant in the original formula was negative.
It is not a role of the present paper to provide a complete
numerical discussion. A comparison between the original
and the corrected formulas clearly justifies the necessity of
the proposed correction in Bahrs phase-deviation tensor-
7. Conclusion
According to numerical tests carried out on synthetic mod-
els, the differences between the two formulas cannot be ne-
glected. Therefore, it is recommended to use this improved
tensor decomposition method on field data.
Acknowledgments. This work was supported by the Hungarian
Research Foundation (contract Nos. T014882 and T029443). Com-
ments by M. Eisel largely contributed to the quality of the paper.
Bahr, K., Interpretation of the magnetotelluric impedance tensor: Regional
induction and local telluric distortion, J. Geophys., 62, 119127, 1988.
Bahr, K., Geological noise in magnetotelluric data: A classification of dis-
tortion types, Phys. Earth Planet. Int., 66, 2438, 1991.
E. Pracser (e-mail: pracser@elgi.hu) and L. Szarka (e-mail: szarka@
ggki.hu)