Beruflich Dokumente
Kultur Dokumente
Razvigor Ossikovski
Oriol Arteaga
Abstract. We describe a numerical method for obtaining a nondepolarizing estimate from an experimental
Mueller matrix, a necessary preliminary step in determining the Jones matrix and the polarization properties
of the sample under study. The proposed method, being a variant of the general virtual experiment approach,
is based on minimizing the least squares distance between the light intensities virtually generated by the effec-
tively measured Mueller matrix of the sample and by its nondepolarizing estimate, while taking into account the
exact phenomenological description of the polarimetric instrument used. It can be applied to complete, as well as
to partial (12-element) experimental Mueller matrices. The application of the method is illustrated on experimen-
tal examples and its performance is compared to that of alternative approaches. © 2019 Society of Photo-Optical
Instrumentation Engineers (SPIE) [DOI: 10.1117/1.OE.58.8.082409]
Keywords: polarimetry; depolarization; Mueller matrix; Jones matrix.
Paper 181692SS received Nov. 28, 2018; accepted for publication Feb. 8, 2019; published online Feb. 26, 2019.
combinations of Mueller matrix elements, one needs to Mueller matrix M is the one that would produce light inten-
modify and extend the approach presented below. sities that are as close as possible to those potentially gen-
erated by M.9,10 If least squares estimation is supposed,9 then
2 Formal Description of the Method one must minimize the normalized residual δ2 given as
A Mueller matrix polarimeter measures the M × N positive 1
matrix I of light intensities in accordance with the fundamen- δ2 ¼
EQ-TARGET;temp:intralink-;e004;326;708 kA M W − A Mnd Wk2
MN
tal measurement relation
1
¼ kAðM − Mnd ÞWk2 ; (4)
I ¼ A M W;
EQ-TARGET;temp:intralink-;e001;63;674 (1) MN
where W and A are the 4 × N and M × 4 matrices of where the notation k: : : k stands for the Frobenius matrix
the polarization state generator (PSG) and the polarization norm.
p (The ffi Frobenius norm of the matrix B is kBk ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
state analyzer (PSA) of the instrument, respectively, M is trðBT BÞ, where “tr” is the matrix trace, so that kBk2
the 4 × 4 Mueller matrix of the sample, and NðMÞ is the equals the sum of the squares of the elements of B.) The min-
number of polarization states generated (analyzed) by the imization parameters are the four complex elements J ij ,
PSG (PSA).2,5 i; j ¼ 1; 2, of the Jones matrix J through which the non-
The two instrument matrices W and A have the respective depolarizing estimate Mnd is expressed as1
forms
Mnd ¼ TðJ ⊗ J ÞT−1 ;
EQ-TARGET;temp:intralink-;e005;326;561 (5)
W¼
EQ-TARGET;temp:intralink-;e002a;63;555 ½ sð1Þ
i
ð2Þ
si :::
ðNÞ
si ; (2a)
where the transition matrix T is given as1
and 21 0 0 1 3
A ¼ ½ sð1Þ
o
ð2Þ
so :::
ðMÞ T
so ; (2b) 61 0 0 −1 7
T¼4
EQ-TARGET;temp:intralink-;e002b;63;510
5: (6)
0 1 1 0
EQ-TARGET;temp:intralink-;e006;326;518
ðkÞ ðkÞ
where si ðso Þ
is the input (output) Stokes vector describing 0 i −i 0
the k’th polarization state generated (analyzed) by the PSG
(PSA). [The superscript T in Eq. (2b) stands for matrix (The asterisk stands for complex conjugation; the symbol
transpose.] “⊗” denotes the Kronecker product.) Since Mnd is real,
The above measurement formalism holds for a Mueller one can assume that one of the elements J ij of J is real
polarimeter employing M × N discrete polarization states
too which reduces the number of real fitting parameters to
(e.g., a ferroelectric liquid crystal based one). If continuous
4 × 2 − 1 ¼ 7. Once the Jones matrix elements are deter-
periodic modulation of the polarization state of light is used
mined by the minimization procedure, the nondepolarizing
instead (e.g., in a rotating compensator or a phase-modulated
instrument), then the PSG and the PSA matrices are time- estimate Mnd is obtained from Eq. (5).
dependent, WðtÞ ¼ ½si ðtÞ and AðtÞ ¼ ½so ðtÞT . However, Clearly, the PSG and PSA matrices W and A depend on
the discrete forms from Eq. (2) can still be used provided the specific design of the instrument. For a discrete state
the continuous periodic signals are sampled in accordance complete Mueller polarimeter, such as a ferroelectric crystal
with the Nyquist criterion based one, one typically has M ¼ N ¼ 4 for the number of
generated and analyzed polarization states. If the Stokes
WðtÞ → W ¼ ½ si ðt1 Þ
EQ-TARGET;temp:intralink-;e003a;63;327 si ðt2 Þ ::: si ðtN Þ ; (3a) vectors sðkÞ of these states, assumed identical for the PSG
and the PSA, are uniformly distributed in polarization
and space, i.e., form a regular tetrahedron on the Poincaré sphere,
then
AðtÞ → A ¼ ½ so ðt1 Þ so ðt2 Þ
EQ-TARGET;temp:intralink-;e003b;63;285 ::: so ðtM Þ T : (3b)
W ¼ AT ¼ ½ sð1Þ sð2Þ sð3Þ sð4Þ
To measure the complete 4 × 4 Mueller matrix M, the
EQ-TARGET;temp:intralink-;e007;326;274
21 1 1 1 3
number of polarization states generated and analyzed by
the instrument must be N ≥ 4 and M ≥ 4. Furthermore, 6 a −a a −a 7
¼4 5; (7)
ðkÞ ðkÞ
the input and output Stokes vectors si and so entering a −a −a a
the instrument matrices W and A from Eq. (2) must be com- a a −a −a
plete, i.e., they must not contain any identically vanishing pffiffiffi
components. The instrument is then termed a complete where a ¼ 1∕ 3. This is the conventional version of the VE
ðkÞ ðkÞ method.9 Since the value of the residual defined by Eq. (4) is
Mueller polarimeter. If N < 4 or M < 4, or if si or so independent of the way the PSG and PSA regular tetrahedra
are incomplete, the Mueller polarimeter is partial. An impor- are oriented in polarization space, Eq. (7) can be used to
tant special case of a partial polarimeter is the generalized
ðkÞ ðkÞ
model any 16-state Mueller polarimeter that is optimal,
ellipsometer where si (so ) contain exactly one vanishing i.e., whose four PSG and four PSA polarization states are
ðkÞ ðkÞ
component whereas so (si ) are complete; such an instru- both uniformly distributed in polarization space, regardless
ment measures 12 out of the 16 elements of the Mueller of their specific orientation. Complete Mueller polarimeters
matrix with a column (a row) missing. based on continuous time modulation, e.g., the dual rotating
The principle of the VE method states that the best non- compensator (DRC)3,5,13 or the four photoelastic modulator
depolarizing estimate Mnd of the experimentally determined (4PEM)14 polarimeters, have time-dependent (and generally
different) PSG and PSA matrices WðtÞ and AðtÞ. If the the PSA are mirror images of each other, then WðtÞ ¼ AðtÞT
arrangements of the optical components in the PSG and in and si ðtÞ ¼ so ðtÞ. For the DRC polarimeter, one then has
T
si ðtÞ ¼ so ðtÞ ¼ 1
EQ-TARGET;temp:intralink-;e008;63;718
cos2 R2C þ sin2 R2C cos 4ωC t sin2 R2C sin 4ωC t sin RC sin 2ωC t ; (8)
where RC and ωC are the retardance and the angular speed of the PSG (PSA) compensator, respectively.5,13 For the 4PEM
polarimeter,
si ðtÞ ¼ so ðtÞ ¼ ½ 1
EQ-TARGET;temp:intralink-;e009;63;650 cos Rm1 ðtÞ sin Rm1 ðtÞ sin Rm2 ðtÞ sin Rm1 ðtÞ cos Rm2 ðtÞ T ; (9)
k − 1 2π ρpp ρps
tk ¼ ; k ¼ 1; 2; : : : ; N: (13) J¼ ; (15)
N ωmax ρsp 1
EQ-TARGET;temp:intralink-;e013;63;752
EQ-TARGET;temp:intralink-;e015;326;752
Thus, for the rotating polarizer PSG and the rotating com-
then Table 1 reports the values of the elements of J obtained
pensator PSG (PSA) ωmax ¼ 2ωP and ωmax ¼ 4ωC ; see
Eqs. (10) and (8), respectively. These are the values used with various methods.
in modeling the RPCE and the RPPME instruments, as The first two lines (header excluded), named “Cloude”
well as the DRC complete polarimeter. For the phase modu- and “VE,” report the Jones matrices of the nondepolarizing
lated PSA in the RPPME instrument ωmax ¼ ωm , whereas estimates obtained with the Cloude decomposition8 and the
for the 4PEM complete polarimeter, ωmax ¼ ωm1 ¼ ωm2 . conventional VE9 methods. The values in the next two lines
The retardance of the compensator(s) was set to result from DRC and 4PEM complete polarimeter VEs. The
RC ¼ π∕4; the modulation amplitude Am of the PSA of last column of the table reports the root mean square (rms)
the RPPME instrument, as well as those of the PSG and deviation ΔM of the matrix elements of the experimental M
the PSA of the 4PEM complete polarimeter, assumed from those of its nondepolarizing estimate Mnd
equal, Am1 ¼ Am2 , were all set to 2.405 rad. By using the sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffi
above values, the W and A matrices of all four instruments 1 X
(RPCE, RPPME, and DRC, and 4PEM polarimeters) were ΔM ¼
EQ-TARGET;temp:intralink-;e016;326;607 M − Mndij ;
i;j ij
(16)
computed. In accordance with the fundamental measurement N
relation Eq. (1), the W and A matrices will generate, in the
case of N ¼ 3 sampling points, a total of N × N ¼ 3 × 3 ¼ 9 in which the sum runs either over all 16 matrix elements in
virtual intensities which is the minimum number making the case of complete polarimetry (and N ¼ 16) or over the 12
possible the determination of the seven-parameter Jones elements belonging to the first three columns in the case of
matrix J from either the complete, 16-element, or the partial, partial polarimetry (generalized ellipsometry; N ¼ 12). It
12-element, experimental Mueller matrix M by using should be mentioned that, in the partial polarimetry case,
Eqs. (4)–(6). the Jones matrix quality factor18 can be used instead of
ΔM. The evaluation of ΔM allows for the assessment of
3.1 First Example the quality of the nondepolarizing estimation. The ΔM val-
ues for all four complete polarimetry methods are compa-
Consider the experimental Mueller matrix rable in magnitude and lie below 0.3%; to be compared
2 1 −0.8351 0.1757 0.1525 3
to the experimental error which has been estimated to be
below 0.5% for the 4PEM polarimeter.14 It should be
6 −0.8245 0.9268 −0.1075 −0.2692 7
M1 ¼ 6
4 −0.1761
7; (14) noted that in the case of the 4PEM VE we had to oversample
−0.5129 0.0424 5 the period in five instead of in only three points, i.e., we set
EQ-TARGET;temp:intralink-;e014;63;438
0.1054
0.1510 −0.2673 −0.0399 −0.4426 N ¼ 5 in Eq. (13), to be able to achieve accuracy comparable
to that of the other three methods.
of a diffraction grating measured in a reflection configuration The last three lines of Table 1 report the Jones matrix esti-
at the wavelength of 540 nm and at the incidence angle of mates obtained from RPCE and RPPME partial polarimetry
65 deg by using a 4PEM polarimeter.14 Details on the struc- VEs. The last but one line, named “RPCE with sym.” gives
ture of the diffraction grating sample can be found in Ref. 17. the Jones matrix with the imposed symmetry condition
The matrix M1 is only very weakly depolarizing; its Gil- ρsp ¼ −ρps ; it was therefore obtained by fitting only five
Bernabeu depolarization index (DI)2,4 equals 0.995 (recall instead of all seven Jones matrix elements, as explained at
that DI ¼ 1 for a nondepolarizing Mueller matrix). the end of the previous section. Indeed, the grating sample
Assuming that the Jones matrix J corresponding to the is invariant with respect to a 180-deg-rotation about its nor-
best nondepolarizing estimate Mnd of the experimental mal and therefore, it should feature an antisymmetric Jones
Mueller matrix M1 is normalized to its J 22 element matrix.
Table 1 Jones matrix elements associated with the best nondepolarizing estimates of M1 obtained with various methods.
RPCE with sym. −0.2683 þ 0.0182i −0.0791 − 0.1146i 0.0791 þ 0.1146i 0.0034
Table 2 Jones matrix elements associated with the best nondepolarizing estimates of M2 obtained with various methods.
RPCE with sym. –0.7993 þ 0.3463i −0.1455 − 0.1560i 0.1455 þ 0.1560i 0.0192
Like in the complete polarimetry case, the partial polar- depolarization metrics, e.g., the depolarization index, are
imetry VEs feature comparable rms deviation values (last based on the complete, 16-element experimental Mueller
column in Table 1). More specifically, the ΔM values in matrix.
the partial polarimetry case are higher than those of the com- Another “effect” of the higher depolarization of M2 is the
plete polarimetry (DRC and 4PEM) ones (by about 40%); larger asymmetry between the two off-diagonal Jones matrix
nevertheless, they are still of the order of the experimental elements ρsp and ρps . Indeed, these are supposed to be oppo-
error of the 4PEM polarimeter (<0.5%). It should be stressed site to one another (ρsp ¼ −ρps ) because of the symmetry
that the neither the Cloude nor the conventional VE properties of the diffraction grating sample (note that the
approaches can be applied to the partial polarimetry case, presence of depolarization does not affect the symmetries).
since both require a complete experimental Mueller matrix. Close inspection of the respective table entries indicates that
the PRCE partial polarimetry VE features larger asymmetry
3.2 Second Example between off-diagonal matrix elements than the complete
polarimetry ones (Cloude, VE, or DRC). Thus, the violation
The experimental Mueller matrix
of the expected symmetry relations between the estimated
2 1 −0.1261 0.2189 −0.0418 3 Jones matrix elements in the partial (12-element) polarimetry
case appears as an extra (indirect) indicator, besides the rms
6 −0.1280 0.8674 −0.0677 −0.3254 7
M2 ¼ 6
4 −0.2140
7 (17) deviation value, for the presence of high depolarization in
−0.8924 0.3744 5
EQ-TARGET;temp:intralink-;e017;63;452