-
Notifications
You must be signed in to change notification settings - Fork 4
/
631019.txt
1098 lines (831 loc) · 22.4 KB
/
631019.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
An Introduction to Complex Differentials and
Complex Differentiability
Raphael Hunger
Technical Report
TUM-LNS-TR-07-06
2007
Technische Universit¨ t M¨ nchen
a u
Associate Institute for Signal Processing
Prof. Dr.-Ing. Wolfgang Utschick
Contents
1. Introduction
3
2. Complex Differentiability and Holomorphic Functions
4
3. Differentials of Analytic and Non-Analytic Functions
8
4. Differentials of Real-Valued Functions
11
5. Derivatives of Functions of Several Complex Variables
14
6. Matrix-Valued Derivatives of Real-Valued Scalar-Fields
17
Bibliography
20
2
1. Introduction
This technical report gives a brief introduction to some elements of complex function theory. First,
general definitions for complex differentiability and holomorphic functions are presented. Since
non-analytic functions are not complex differentiable, the concept of differentials is explained both
for complex-valued and real-valued mappings. Finally, multivariate differentials and Wirtinger
derivatives are investigated.
3
2. Complex Differentiability and Holomorphic Functions
Complex differentiability is defined as follows, cf. [Schmieder, 1993, Palka, 1991]:
Definition 2.0.1. Let A ⊂ C be an open set. The function f : A → C is said to be (complex)
differentiable at z0 ∈ A if the limit
f (z) − f (z0 )
lim
(2.1)
z→z0
z − z0
exists independent of the manner in which z → z0 . This limit is then denoted by f (z0 ) =
and is called the derivative of f with respect to z at the point z0 .
df (z)
dz z=z0
A similar expression for (2.1) known from real analysis reads as
df (z)
f (z + ∆z) − f (z)
= lim
,
∆z→0
dz
∆z
(2.2)
where ∆z ∈ C now holds. Note that if f is differentiable at z0 then f is continuous at z0 . An
equivalent, but geometrically more illuminating way to define the derivative follows from the linear
approximation of f in the local vicinity of z0 [Palka, 1991].
Definition 2.0.2. Let A be an open set. The function f : A → C is said to be (complex) differentiable at z0 ∈ A if there exists a complex-valued scalar g such that
f (z) = f (z0 ) + g · (z − z0 ) + e(z, z0 ),
(2.3)
holds for every z ∈ A and the function e(·, ·) satisfies the condition
lim
z→z0
e(z, z0 )
= 0.
z − z0
4
(2.4)
2. Complex Differentiability and Holomorphic Functions
5
The remainder term e(z, z0 ) in (2.4) obviously is o(|z − z0 |) for z → z0 and therefore g · (z − z0 )
dominates e(z, z0 ) in the immediate vicinity of z0 if g = 0. Close to z0 , the differentiable function
f (z) can linearly be approximated by f (z0 ) + f (z0 )(z − z0 ). The difference z − z0 is rotated by
∠f (z0 ), scaled by |f (z0 )| and afterwards shifted by f (z0 ).
The concept of a differentiability in a single point readily extends to differentiability in open sets.
Definition 2.0.3. Let U ⊆ A be a nonempty open set. The function f : A → C is called holomorphic (or analytic) in U, if f is differentiable in z0 for all z0 ∈ U. Moreover, if f is analytic in the
complete open domain-set A, f is a holomorphic (analytic) function.
An interesting characteristic of a function f analytic in U is the fact that its derivative f is analytic
in U itself [Spiegel, 1974]. By induction, it can be shown that derivatives of all orders exist and
are analytic in U which is in contrast to real-valued functions, where continuous derivatives need
not be differentiable in general. However, basic properties for the derivative of a sum, product, and
composition of two functions known from real-valued analysis remain inherently valid in the complex domain. Assume that f (z) and g(z) are differentiable at z0 . Then, the following propositions
hold:
Proposition 2.0.1. The sum f + g is differentiable at z0 and
(f + g) (z0 ) = f (z0 ) + g (z0 ).
(2.5)
Proposition 2.0.2. The product f g is differentiable at z0 and
(f g) (z0 ) = f (z0 )g(z0 ) + f (z0 )g (z0 ).
Proposition 2.0.3. If g(z0 ) = 0, the quotient
f
g
(z0 ) =
f
g
(2.6)
is differentiable at z0 and
f (z0 )g(z0 ) − f (z0 )g (z0 )
.
g 2 (z0 )
(2.7)
Proposition 2.0.4. If f is differentiable at g(z0 ), the composition f ◦ g is differentiable at z0 and
(f ◦ g) (z0 ) = f (g(z0 ))g (z0 ) (chain rule).
(2.8)
Complex differentiability is closely related to the Cauchy-Riemann equations [Lang, 1993]. A necessary condition for f being holomorphic in U requires the Cauchy-Riemann equations to be satisfied.
6
2. Complex Differentiability and Holomorphic Functions
Theorem 2.0.1: Let f (z) = u(z) + jv(z) with u(z), v(z) ∈ R and z = x + jy with x, y ∈ R.
In terms of x and y, the function f (z) can be expressed as F (x, y) = U (x, y) + jV (x, y) with
U (x, y), V (x, y) ∈ R. A necessary condition for f (z) being holomorphic in U is that the following
system of partial differential equations termed Cauchy-Riemann-equations holds for every z =
x + jy ∈ U:
∂V (x, y)
∂U (x, y)
∂V (x, y)
∂U (x, y)
=
and
=−
.
(2.9)
∂x
∂y
∂y
∂x
Proof : According to Definition 2.0.3, f (z) is holomorphic in U if f (z) is differentiable at every
z ∈ U. Differentiability at z implies that the limit
f (z + ∆z) − f (z)
∆z→0
∆z
= lim
lim
z=x+jy
∆x→0
∆y→0
F (x + ∆x, y + ∆y) − F (x, y)
∆x + j∆y
exists no matter which curve ∆z moves along when approaching zero, see Definition 2.0.1
and (2.2). Setting ∆z = ∆x + j∆y, two possible curves for ∆z → 0 are considered. The first
curve goes in the horizontal direction with ∆y = 0 and ∆x → 0 yielding
F (x + ∆x, y) − F (x, y)
∆x→0
∆x
U (x + ∆x, y) − U (x, y)
V (x + ∆x, y) − V (x, y)
+j
= lim
∆x→0
∆x
∆x
∂V (x, y)
∂U (x, y)
+j
.
=
∂x
∂x
The second curve goes in the vertical direction with ∆x = 0 and ∆y → 0 yielding
f (z = x + jy) = lim
F (x, y + ∆y) − F (x, y)
∆y→0
j∆y
V (x, y + ∆y) − V (x, y)
U (x, y + ∆y) − U (x, y)
+j
= lim
∆y→0
j∆y
j∆y
∂U (x, y) ∂V (x, y)
+
.
=
j∂y
∂y
As both expressions have to be the same for f (z) being holomorphic, (2.9) immediately follows
as a necessary condition.
f (z = x + jy) = lim
2
The next theorem provides conditions under which the Cauchy-Riemann equations are sufficient
for f (z) being holomorphic.
Theorem 2.0.2: If the partial derivatives of U (x, y) and V (x, y) with respect to x and y are continuous, the Cauchy-Riemann equations are sufficient for f (z) being holomorphic.
Proof :
See [Spiegel, 1974].
2
2. Complex Differentiability and Holomorphic Functions
In the following, we give examples for analytic functions and functions which are not analytic.
Examples for analytic functions:
•
f (z) = z n
•
f (z) =
•
f (z) = ln(z)
•
f (z) = exp(az)
az + b
cz + d
f (z) = nz n−1
ad − bc
(cz + d)2
1
f (z) =
z
f (z) =
f (z) = a exp(az)
Examples for non-analytic functions:
• f (z) = |z|2
• f (z) = {z}
• f (z) = {z}
• f (z) = z ∗
7
3. Differentials of Analytic and Non-Analytic Functions
The total differential of the bivariate function F (x, y) associated to the univariate function f (z)
via F (x, y) = U (x, y) + jV (x, y) = f (z)|z=x+jy reads as [Henrici, 1974]
dF =
∂F (x, y)
∂F (x, y)
dx +
dy.
∂x
∂y
(3.1)
Of course, differentiability of F (x, y) with respect to x and y in the real sense has to be imposed
for the existence of the differential dF in (3.1). This implies the differentiability of the real-valued
functions U (x, y) and V (x, y) with respect to x and y. Rewriting (3.1) by means of F (x, y) =
U (x, y) + jV (x, y) yields
dF =
∂U (x, y)
∂V (x, y)
∂U (x, y)
∂V (x, y)
dx + j
dx +
dy + j
dy.
∂x
∂x
∂y
∂y
(3.2)
Making use of
dz = dx + jdy,
dz ∗ = dx − jdy,
(3.3)
the two differentials dx and dy can be expressed via
1
(dz + dz ∗ )
2
1
dy = (dz − dz ∗ ) .
2j
dx =
(3.4)
Inserting (3.4) into the differential expression dF in (3.1) and reordering the result leads to
dF =
1 ∂U (x, y) ∂V (x, y)
∂V (x, y) ∂U (x, y)
+
+j
−
dz
2
∂x
∂y
∂x
∂y
1 ∂U (x, y) ∂V (x, y)
∂V (x, y) ∂U (x, y)
+
dz ∗ .
−
+j
+
2
∂x
∂y
∂x
∂y
(3.5)
A major result can already be anticipated here.
Proposition 3.0.1. The differential of any analytical function f (z) does not depend on the differential dz ∗ .
8
3. Differentials of Analytic and Non-Analytic Functions
9
Proof : Since any analytical function f (z) satisfies the Cauchy-Riemann equations in (2.9), the
factor in front of dz ∗ in the second line of (3.5) is zero. Obviously, the differential dF does not
depend on dz ∗ .
2
Note that the converse of Proposition 3.0.1 is also true: If the differential of a function f does not
depend on dz ∗ , the function f is analytical.
Rearranging the terms in (3.5), the differential dF can be expressed as
dF =
∂
1 ∂
(U (x, y) + jV (x, y)) − j (U (x, y) + jV (x, y)) dz
2 ∂x
∂y
1 ∂
∂
+
(U (x, y) + jV (x, y)) + j (U (x, y) + jV (x, y)) dz ∗ .
2 ∂x
∂y
Recognizing that U (x, y) + jV (x, y) = F (x, y), we finally obtain by factoring out the partial
differential operators
dF =
∂
∂
1 ∂
1 ∂
−j
+j
F (x, y) dz +
F (x, y) dz ∗ .
2 ∂x
∂y
2 ∂x
∂y
(3.6)
According to the total differential for real-valued multivariate functions, the introduction of the
∂
∂
two operators ∂z and ∂z ∗ is reasonable as it leads to the very nice description of the differential df ,
where the real-valued partial derivatives are hidden [Trapp, 1996].
Theorem 3.0.1: The differential df of a complex-valued function f (z) : A → C with A ⊂ C
can be expressed as
∂f (z)
∂f (z) ∗
df =
dz .
(3.7)
dz +
∂z
∂z ∗
Proof : See the preceding derivation and the definition of the partial derivatives
following.
∂
∂z
and
∂
∂z ∗
in the
2
Definition 3.0.1. The two ‘partial derivative’ operators
∂
∂z
and
1 ∂
∂
∂
:=
−j
,
∂z
2 ∂x
∂y
∂
∂
1 ∂
+j
:=
,
∗
∂z
2 ∂x
∂y
∂
∂z ∗
are defined by
(3.8)
and are often referred to as the Wirtinger derivatives [Wirtinger, 1926] and correspond to half of
the del-bar and del operator [Spiegel, 1974].
10
3. Differentials of Analytic and Non-Analytic Functions
Basic rules for this Wirtinger calculus are given in the next theorem.
Theorem 3.0.2: For the Wirtinger derivatives, the common rules for differentiation known from
real-valued analysis concerning the sum, product, and composition of two functions hold as well.
In particular,
∂ ∗
∂
z = ∗ z = 0,
∂z
∂z
which means that z ∗ can be regarded as a constant when differentiating with respect to z, as well
as z can be regarded constant when differentiating with respect to z ∗ .
Proof :
With z = x + jy and z ∗ = x − jy, Theorem 3.0.2 follows immediately from (3.8).
2
Examples:
•
∂ 2
∂
|z| =
(zz ∗ ) = z ∗
∂z
∂z
•
∂
∂
exp −|z|2 =
exp (−zz ∗ ) = −z ∗ exp |z|2
∂z
∂z
Corollary 3.0.1. Derivatives of the conjugate function f ∗ (z) satisfy the relationships
∂f ∗ (z)
=
∂z
Proof :
∂f (z)
∂z ∗
∗
and
∂f ∗ (z)
=
∂z ∗
See the definition of the Wirtinger derivatives in (3.8).
∂f (z)
∂z
∗
.
(3.10)
2
4. Differentials of Real-Valued Functions
Optimizations in communications and signal processing are frequently targeted on the maximization of a utility or on the minimization of a cost. Hence, most objectives are real-valued, as the
standard total order can only handle real-valued arguments. On account of this, this chapter deals
with functions f (z) : U → R having complex-valued arguments z ∈ U ⊂ C that are mapped
to real-valued scalars f (z) ∈ R. In addition, simplifications resulting from this circumstance are
investigated.
First of all it is obvious that the only possibility of a real-valued function f (z) with complex
argument z for being analytic is that f (z) is constant for all z of its domain. This follows from the
Cauchy-Riemann equations in (2.9), since v(z) = V (x, y) = {f (z)} = 0 for real-valued f (z).
This leads to the following proposition.
Proposition 4.0.1. All non-trivial (not constant) real-valued functions f (z) mapping z ∈ A ⊂ C
onto R are non-analytic functions and therefore not complex differentiable.
With the definition of the differential dF in (3.2) and (3.6), it is easy to prove the following theorem.
Theorem 4.0.1: The differential df of a real-valued function f (z) : A → R with complex-valued
argument z ∈ A ⊂ C can be expressed as
df = 2
∂f (z)
dz
∂z
=2
∂f (z) ∗
dz
∂z ∗
(4.1)
and is equivalent to
dF =
∂F (x, y)
∂F (x, y)
dx +
dy.
∂x
∂y
(4.2)
Due to the property that the non-trivial real-valued functions are not analytic, stationary points of
the real-valued f (z) cannot be obtained by searching for points z where the derivative f (z) is
zero. However, we can detect stationary points z of f (z) by a vanishing differential df .
11
12
4. Differentials of Real-Valued Functions
Theorem 4.0.2: The differential df of a real-valued function f (z) : A → R with complex
argument z ∈ A ⊂ C vanishes if and only if the Wirtinger derivative is zero:
df = 0 ⇔
∂f (z)
= 0.
∂z
(4.3)
Proof : First, we prove that ∂f (z) = 0 leads to df = 0. This is a result from (4.1). The converse is
∂z
shown by the following reasoning. For arbitrary ratios of dx and dy, dF from (4.2) and therefore
df can only vanish if both partial derivatives of F (x, y) with respect to x and y are zero. With
∂f (z)
1 ∂U (x, y)
∂U (x, y)
(dx + jdy) ,
dz =
−j
∂z
2
∂x
∂y
(4.4)
vanishing partial derivatives document the second way of the equivalence relation in Theorem 4.0.2
as F (x, y) = U (x, y) for real-valued f (z).
2
Gradient-based iterative algorithms targeted on maximizing or minimizing an objective function
can be constructed by optimizing dz in the differential expression (4.1).
Corollary 4.0.1. The steepest ascent of a real-valued function f (z) : A → R with complexvalued argument z ∈ A ⊂ C is obtained for
dz =
∂f (z)
ds,
∂z ∗
where ds is a real-valued differential. Thus, the steepest ascent points to the direction of
Proof :
(4.5)
∂f (z)
.
∂z ∗
As the differential df of a real-valued function can be expressed as (see Equ. 4.1)
df = 2
∂f (z) ∗
dz ,
∂z ∗
df is maximized for real-valued ∂f (z) dz ∗ if the norm of dz is fixed. Hence, dz ∗ has to be a scaled
∂z ∗
∂f (z)
version of the conjugate of ∂z ∗ . Equivalently, dz must be a scaled version of ∂f (z) and (4.5)
∂z ∗
immediately follows.
2
An iterative implementation could therefore read as
z ←z+2
∂f (z)
ds,
∂z ∗
4. Differentials of Real-Valued Functions
13
where ds can be interpreted as the step-size. Notice that there is a factor 2 in front of the Wirtinger
derivative which follows from (4.1)!
5. Derivatives of Functions of Several Complex Variables
When switching to functions of several complex variables stacked in the column vector z =
[z1 , . . . , zn ]T ∈ Cn , we confine ourselves to mappings onto the one-dimensional complex domain C. For them, the holomorphic-property is defined by the following two definitions which are
equivalent [Krantz, 1992]:
Definition 5.0.1. A function f (z) : Cn ⊃ A → C is said to be holomorphic if for each k ∈
{1, . . . , n} and each fixed z1 , . . . , zk−1 , zk+1 , . . . , zn the function
w → f ([z1 , . . . , zk−1 , w, zk+1 , . . . , zn ]T )
is holomorphic according to the one-dimensional sense in Definition 2.0.3 on the set {w ∈ C :
[z1 , . . . , zk−1 , w, zk+1 , . . . , zn ]T ∈ A}.
This is nothing else than that the function f (z) has to be holomorphic in each variable z 1 , . . . , zn .
An equivalent definition reads as follows [Krantz, 1992]:
Definition 5.0.2. A function f (z) : A → C that is continuously differentiable in each variable z k ,
k ∈ {1, . . . , n} is said to be holomorphic if the Cauchy-Riemann equations are satisfied in each
variable separately.
Although there are many differences between univariate and multivariate complex functions, the
Wirtinger calculus easily extends to the case of several complex variables.
Theorem 5.0.1: The differential df of a multivariate complex-valued function f (z) : A → C
with A ⊂ Cn can be expressed as
n
df =
k=1
∂f (z)
dzk +
∂zk
n
k=1
∂f (z) ∗
dzk
∗
∂zk
∂f (z)
∂f (z) ∗
=
dz +
dz .
T
∂z
∂z H
14
(5.2)
5. Derivatives of Functions of Several Complex Variables
Note that the operators
∂
∂z
and
∂
∂z ∗
15
read as
∂
∂
∂
=
,...,
∂z
∂z1
∂zn
T
∂
∂
∂
=
,..., ∗
∗
∗
∂z
∂z1
∂zn
T
,
(5.3)
.
If they are applied to a scalar field f (z) they generate a column vector of dimension n and mimic
the gradient operator for real-valued functions. Again, both the differential dz = [dz 1 , . . . , dzn ]T
and its conjugate dz ∗ are required to express the differential df in (5.2) for non-analytic functions.
Similar to Theorem 4.0.2, we make the following observation for mappings onto R:
Theorem 5.0.2: The differential df of a real-valued function f (z) : A → R with complex
argument z ∈ A ⊂ Cn vanishes if and only if the vector-valued Wirtinger derivative is zero:
∂f (z)
= 0.
∂z
df = 0 ⇔
(5.4)
Finally, the conjugate Wirtinger derivative again points into the direction of the steepest ascent:
Corollary 5.0.1. The steepest ascent of a real-valued function f (z) : A → R with complexvalued argument z ∈ A ⊂ Cn is obtained for
dz =
∂f (z)
ds,
∂z ∗
(5.5)
where ds is a real-valued differential.
Proof :
From Theorem 5.0.1, the differential of the function f with real-valued image reads as
df = 2
∂f (z)
dz .
∂z T
According to the Cauchy-Schwarz inequality, dz has to be the conjugate of
valued differential and the proof is complete.
∂f (z) T
∂z T
times a real2
A gradient ascent step could for example read as
z ←z+2
with the step-size ds.
∂f (z)
ds
∂z ∗
(5.7)
16
5. Derivatives of Functions of Several Complex Variables
Examples for vector-valued Wirtinger derivatives:
∂f (z)
= AT z ∗
• f (z) = z H Az
∂z
∂f (z)
= zHA
∂z T
∂f (z)
= Az
∂z ∗
∂f (z)
= z T AT
H
∂z
6. Matrix-Valued Derivatives of Real-Valued Scalar-Fields
In this section, derivatives of real-valued scalar-fields are investigated. Common representatives of
such scalar fields in communications and signal processing are trace or determinant expressions.
Definition 6.0.1. Let f : Cn×n → R denote a functional acting as a map A → f (A). Then the
derivative of f (A) with respect to the matrix A returns a matrix-valued function whose entry in
the m-th row and n-th column reads as
∂
f (A)
∂A
=
m,n
∂
f (A).
∂[A]m,n
(6.1)
Sometimes the functional f is a composition of several operations where the outer one is a linear
operator like the trace-operator for example. In such a case, the partial derivative operator and this
outer linear operator may be interchanged as their commutator vanishes:
∂
∂
tr(A(t)) = tr
A(t) .
∂t
∂t
Clearly, derivatives of matrices with respect to scalars turn then out to be necessary and will therefore be discussed now.
The derivative of a matrix A(t) with respect to the variable t the matrix depends on follows from
the element-wise application of the partial derivative operator onto the entries of A. Hence, the
element in the m-th row and n-th column of the derivative reads as
∂
A(t)
∂t
=
m,n
∂
[A(t)]m,n .
∂t
(6.3)
If t is complex valued, then the partial derivative operator denotes the Wirtinger derivative. Equivalently, t may stand for an element of the matrix A(t). For example, if t = [A] m,n = am,n we
have
∂
A = e m eT ,
(6.4)
n
∂am,n
where em denotes the m-th canonical unit vector of appropriate dimension the elements of which
are all zero except for the one in the m-th row. For the following examples, all matrices are assumed
17
18
6. Matrix-Valued Derivatives of Real-Valued Scalar-Fields
to be constant and mutually independent. Moreover, no special structure or symmetry is assumed
for them.
Examples:
n
n
n
n
∂A
∂f (A)
T
=
=
• f (A) = tr(A)
ek e tr
ek eT tr ek eT
∂A
∂[A]k,
k=1 =1
k=1 =1
n
n
n
T T
=
k=1 =1
•
f (A) = tr(AB)
ek eT = I n
k
ek e e ek =
∂f (A)
=
∂A
n
k=1
n
∂AB
ek e tr
∂[A]k,
=1
n
T
k=1
n
n
ek eT tr ek eT B
=
n
n
k=1 =1
ek eT eT Bek =
=
k=1 =1
n
ek eT [B]l,k = B T
k=1 =1
Many information theoretic expressions involve the determinant-operator. For the derivative of
them, the following proposition holds: