Complex Derivative: A Basis Free Definition PARTHA PRATIM DEY Department of Computer Science North South University 12 Kemal Ataturk Avenue, Banani, Dhaka BANGLADESH Abstract:- Complex variables as we know is very representation dependent, which many think it need not be. Text books define complex numbers as x iy using basis {1, i} which makes everything including complex derivative, Cauchy-Riemann conditions etc. tied to that particular basis though {1, i} is not the only basis of C . In this paper, we use Frechet derivative to construct a definition of complex derivative, which is basis free in the sense that no basis, either explicit or implicit, has been used in the definition. Key-Words:- Linear operator, matrix, basis, complex, derivative, Frechet derivative. 1 Introduction Let R ( x, y) | x, y R} be a vector space over R with usual addition and multiplication, specifically 2 ( x1 , y1 ) ( x2 , y2 ) ( x1 x2 , y1 y2 ) and ( x y ) x y where R . We define a multiplication between the vectors ( x1 , y1 ), ( x2 , y 2 ) R 2 as follows: ( x1 , y1 ) ( x2 , y2 ) ( x1 x2 y1 y2 , y1 x2 x1 y2 ) The vetor space R 2 equipped with the multiplication above is a field denoted by C and is called the field of complex numbers. Hence a complex number z ( x, y ) is an ordered pair where x and y are called real and imaginary parts of z and denoted by Re z and Im z respectively. Notice that we have not used any basis to define our complex number. Let now B be a basis for vector space R 2 and T be a R -linear operator from R 2 to R 2 with T (a1 ) 1a1 2 a2 and T (a2 ) 1a1 2 a2 . For any v xa1 ya2 in R 2 , let [v] B denote x y , the transpose of the coordinate matrix v relative to basis B and let [T ] B be the matrix such that the j th column of [T ] B is [T ( a j )] B i.e., [T ] B 1 1 . 2 2 [T ] B is regarded as a representation of operator T and is called the matrix of T with respect to basis B . For such a matrix [T (v)] B [T ] B [v] B for any v in R 2 , as T (v ) xT (a1 ) yT (a 2 ) x( 1 a1 2 a 2 ) y ( 1 a1 2 a 2 ) ( x 1 y1 )a1 ( x 2 y 2 )a 2 and x1 y1 1 1 x x y y . 2 2 2 2 Next we quote a well-known theorem [ 1 ] that will be used later to relate matrices of an operator T with respect to various bases. The sketch of the proof given below is ours. Theorem (1.1) Let T be a R -linear operator from R 2 to R 2 . If B ' {a, b} is a basis of R 2 with a (a1 , a2 ) and b (b1 , b2 ) , then [T ] B' P 1[T ] B P where a b P 1 1 a 2 b2 and B is the standard basis comprising of e1 (1,0), e2 (0,1) . Proof. Notice that the equations below a a1e1 a2 e2 b b1e1 b2 e2 is equivalent to the following matrix equation 1 a b1 [e1 , e2 ] [a, b] 1 . a2 b2 Assume now that T (e1 ) 1e1 2 e2 and T (e2 ) 1e1 2 e2 . Then T (a ) T (a1e1 a2 e2 ) 1 1 a1 e1 e2 2 2 a 2 1 1 1 a1 a 2 2 2 a [ a b P 1 [T ] B 1 . a 2 a b1 [ a b 1 a 2 b2 Similarly one can show that b T (b) [a bP 1[T ] B 1 . b2 1 Hence [T ] B' P [T ] B P .■ 2 Matrix Form of a Multiplication Transformation Let M [w] denote the R - linear transformation from C to C given by M [ w]( z ) wz where w, z C and is called a multiplication transformation. The collection M [w] | w Cis a subset of L (C , C ) and is denoted by C * . Since M [ w](1,0) (r , s)(1,0) (r , s) r (1,0) s(0,1) and M [ w](0,1) (r , s )(0,1) ( s, r ) s (1,0) r (0,1) the matrix of M [w] with respect to the canonical basis B {(1,0), (0,1)} has the following form r s [ M ( w)] B s r whereas by Theorem (1.1) , the matrix of M [w] with respect to an arbitrary basis B ' {( a1 , a 2 ), (b1 , b2 )} resembles the following a b1 [ M [ w]] B ' 1 a 2 b2 1 r s s r a1 b1 a b . 2 2 Now we prove a theorem, which states given two bases of C when the rings of matrices of multiplication transformations are the same. Theorem ( 2.1) Let [C * ] B and [C * ] B ' denote the rings of matrices of multiplication transformations with respect to bases B and B ' respectively, where B is the canonical basis {(1,0), (0,1)} and B ' is any other. Then [C * ] B = [C * ] B ' if and only if one of the bases can be obtained from the other by multiplication with a complex number. Proof.Let B {(1,0), (0,1)} and B ' {( x1 , y1 ), ( y1 , x1 )} be two bases of C where elements of B ' are obtained by multiplication of each element of B by ( x1 , y1 ) i.e., ( x1 , y1 ) ( x1 , y1 ) (1,0) and ( y1 , x1 ) ( x1 , y1 ) (0,1) . Let M ( w)[ B ] and M ( w)[ B ' ] be representations of w (r , s ) with respect to bases B and B ' respectively. Then M ( w)[ B ' ] P 1 M ( w)[ B ] P where x y1 P 1 . y1 x1 Since r s , s r M (w)[ B ] we get P 1 M ( w)[ B ] P 1 0 0 1 P 1 r P P 1 s P 0 0 1 1 y1 0 1 x1 1 0 s r 2 . 2 0 1 x1 y1 y1 x1 1 0 x1 y1 y1 x1 . 0 ( x12 y12 ) 1 0 s r 2 2 2 2 0 1 x1 y1 x1 y1 0 r s . s r Assume B ' to be the basis {( x1 , y1 ), ( x2 , y2 )} where ( x2 , y2 ) (1,0) ( x2 , y2 ) , ( x1 , y1 ) (0,1) ( x2 , y2 ) Then x2 y1 and y 2 x1 . Thus B ' can be rewritten as {( x1 , y1 ), ( y1 , x1 )} . Hence P becomes x1 y1 y x . 1 1 Then r [ M ( w)][ B ' ] P s 1 0 0 = P 1 [ r + s 0 1 1 1 0 0 r sP 1 0 1 1 1 1 0 r s 0 1 The equalities above are simplified to: x1 x2 y1 y 2 0 2 2 2 2 x1 y1 x2 y 2 Solving, we get x2 y1 , x1 y2 or x2 y1 , x1 y2 , which shows B ' is a multiplication of B by a complex s P r number. ■ 1 ]P 0 Then [C ] B1 [C ] B2 if and only if one of the bases Theorem ( 2.2) Let B1 , B2 be two bases of C . can be obtained from the other by multiplication with a complex number. Proof. Let B1 ( x1 , y1 ), ( x1' , y1' ) be a basis 1 P 0 0 1 r 1 0 s s . r We now prove the converse. Let B {(1,0), (0,1)} and B ' {( x1 , y1 ), ( x2 , y 2 )} b e two bases of C and [C * ] B [C * ][ B ' ] . We want to show that B {( x1 , y1 ), ( x2 , y 2 )} can be obtained from B {(1,0), (0,1)} by multiplication of each of its element with a complex number, which amounts to one of the following: a. (0,1) ( x1 , y1 ) ( x2 , y2 ) i.e., ' ( y1 , x1 ) ( x2 , y2 ) or equivalently x2 y1 and y 2 x1 b. (0,1) ( x2 , y2 ) ( x1 , y1 ) i.e., ( y2 , x2 ) ( x1 , y1 ) or equivalently x2 y1 and y 2 x1 . Let w (a, b) C and [M (w)] B , [ M ( w)] B ' be the matrix representations of w in bases B and B ' respectively. Then [ M ( w)] B ' [C * ] B ' and hence [ M ( w)] B ' P 1[M ( w)] B P . Since [C * ] B [C * ] B' , we obtain P 1[ M ( w)] B P = [ M ( w)] B ' [C] B , which implies the following. a b( x12 y12 ) b( x22 y 22 ) . x1 y 2 y1 x 2 x1 y 2 y1 x2 b( y1 y 2 x1 x2 ) b( x1 x2 y1 y 2 ) , a x1 y 2 y1 x2 x1 y 2 y1 x2 of C and assume B2 be another where elements of B2 are obtained by multiplication of each element of B1 with an element w (a, b) i.e., B2 {( ax1 by1 , bx1 ay1 ), (ax1' by1' , bx1' ay1' )} Notice that ' ' ax by1 ax1 by1 P2 1 ' ' bx1 ay1 bx1 ay1 ' a b x1 x1 = . ' b a y1 y1 Then r P s 1 2 ' s x1 x1 P2 ' r y1 y1 r s a . s r b b a 1 1 a b 1 b . a ' x1 x1 y ' 1 y1 1 ' ' x x1 r s x1 x1 1 ' ' y1 y1 s r y1 y1 r s P11 P1 . s r Hence [C ] B1 [C ] B2 . We now prove the converse. Suppose [C ] B1 [C ] B2 . Then for any r2 , s2 R , there exist r1 , s1 R such that r P 1 s1 1 1 s1 r s 2 P1 P21 2 P2 r1 s 2 r2 or equivalently, r s1 r2 s 2 1 ( P1 P21 ) 1 1 P P 1 2 s r . s1 r1 2 2 This implies that [C ] B [C ] B' where B is the canonical basis and B ' comprises of the transposed first and second columns of P1 P21 . Then by the Theorem ( 2.2) , there is a complex number (a, b) such that, 1 1 2 ( first column of P P ) transposed (a, b)(1,0) and ( second column of P1 P21 ) transposed (a, b)(0,1) . This gives a b P1 P21 b a or a b P1 P2 , b a which by the discussion in the very beginning of the first part of this proof means the vectors of basis B1 can be obtained from B2 by multiplication of each of its elements with complex number (a, b) .■ 3 Matrix Form of Frechet Derivative Next we discuss a derivative known as the Frechet derivative. Since our definition of complex derivative uses the notion of Frechet derivative, we will say below a few words about Frechet derivative. Definition (3.1) . Let be a norm and E be an open set in R 2 with respect to this norm. Suppose f maps E into R 2 , and z E . If there exists a linear transformation A of R 2 into R 2 such that f ( z h) f ( z ) Ah r (h) where lim r ( h) h 0 h0 then we say that f is differentiable (or Frechet differentiable) at z , and we write f ( z) A . If f is differentiable at every z E , we say that f is differentiable in E . We consider f that maps an open set E R 2 into R 2 and is differentiable at z E . Let B {e1 , e2 } be the canonical basis of R 2 , where as usual e1 (1,0) and e2 (0,1) . The components of f are real functions u, v on E defined by f ( z) u( z)e1 v( z)e2 . It is well-known [ 2 ] that f * ( z )e1 u v ( z )e1 ( z )e2 x x and u v ( z )e1 ( z )e2 . y y Thus by Theorem (1.1) , the Frechet derivative with f * ( z )e 2 respect to canonical basis assumes the following form: (1) u x ( z ) [ f * ( z )] B v ( z ) x u ( z) y . v ( z) y We will use this important fact to prove our next theorem. Theorem (3.2) Let B ' {a, b} be a basis for C over R and f ( z ) u ( x, y )a v( x, y )b be a function from an open set U C to C where z xa yb with a (a1 , a2 ), b (b1 , b2 ) . If f has a Frechet derivative f (z ) at z xa yb , then u x ( x, y) u y ( x, y ) [ f ( z )] B' v x ( x, y) v y ( x, y ) where [ f ( z)] B' is the matrix of Frechet derivative f (z ) with respect to basis B ' . Proof. Set w1 xa1 yb1 , w2 xa2 yb2 . Then w1b2 w2 b1 w1a 2 w2 a1 and y . a1b2 a 2 b1 a1b2 a 2 b1 Now f ( z ) u ( x, y )a v( x, y )b can be rewritten ( 2) x as f (w1 , w2 ) (a1u( x, y), a2u( x, y)) (v( x, y)b1 , v( x, y)b2 ) (a1u ( x, y ) b1v( x, y ))e1 (a 2 u ( x, y ) b2 v( x, y )) where e1 (1,0) and e2 (0,1) constitute the canonical basis B of R 2 and x, y are as in equation ( 2) above. The Frechet derivative of f (w1 , w2 ) with respect to basis B denoted by [ f ( w1 , w2 )] B is then easily computed by (1) . [ f ( w1 , w2 )] B w [a1u b1v] w [a1u b1v] 1 2 w [a 2 u b2 v] w [a 2 u b2 v] 2 1 v u a1 w b1 w 1 1 v u a 2 w b2 w 1 1 a 1 a 2 u v b1 w2 w2 u v a2 b2 w2 w2 a1 u w2 v w2 u b1 w1 b2 v w 1 u a b x 1 1 a 2 b2 v x u 1 y a1 b1 v a 2 b2 y using chain rule [2] in the last step. Setting a b P 1 1, a 2 b2 Now we are ready for a definition of complex derivative. Definition ( 4.1) Let f be a map from an open set E C to C and f be it's Frechet derivative on E . We say that f has a complex derivative f ' ( z ) (or f is C -differentiable) at an element z in E if f * ( z ) C * i.e. f * ( z ) M ( w) for some w C . We let w M 1 ( f * ( z )) to be f ' ( z) . Recall that in the Introduction, we did not use any basis when we defined the complex numbers. Now notice that our definition of complex derivative does not use a basis either. Nor do we use any specific norm, as the definition of Frechet derivative does not use any. Recall here that all norms are equivalent in a finite dimensional vector space and which is what makes the existence of Frechet derivative independent of any norm. The next theorem states the necessary and sufficient conditions of complex differentiability of a complex function f at a complex number z given an arbitrary basis in C . Theorem ( 4.2) Suppose B ' {a, b} is a basis of C over R where a (a1 , a2 ) and b (b1 , b2 ) . Let f be a function from an open set E C to C and is given by f ( z ) u ( x, y )a v( x, y )b when z xa yb E . Then f has a complex derivative f ' ( z 0 ) at z 0 x0 a y0 b if and only if we thus obtain u x ( x, y ) [ f ( w1 , w2 )] B P v ( x, y ) x 4 A New Definition of Complex Derivative u ( x, y ) y P 1 . v ( x, y ) y u x ( x0 , y 0 ) v ( x , y ) x 0 0 u ( x0 , y 0 ) y r s P 1 P v s r ( x0 , y 0 ) y Since by Theorem (1.1), for some r , s R . Moreover f ' ( z 0 ) (r , s ) . [ f (w1 , w2 )] B' P 1[ f (w1 , w2 )] B P , we obtain Proof. Suppose f ' ( z 0 ) exists. Then the desired equality. ■ f ( z 0 ) M ( w) for some w (r , s ) C and f ' ( z 0 ) (r , s ) . Notice that f ( z 0 ) M ( w) implies [ f ( z )] B [M (w)] B . By the last paragraph of Theorem (3.2) , this amounts to: u u ( x0 , y 0 ) ( x , y ) 0 0 x y P 1 P v v ( x , y ) ( x0 , y 0 ) x 0 0 y r s s . r To get the converse we just go backward. ■ Corollary ( 4.3) . ( Cauchy-Riemann Equations) Let f ( z ) u ( x, y )1 v( x, y )i define a function from C to C with the canonical basis B 1, iwhere 1 (1,0) and i (0,1) . Then f has a complex derivative at z 0 x0 1 y 0 i if and only if u v ( x0 , y 0 ) ( x0 , y 0 ) x y and v u ( x0 , y 0 ) ( x0 , y 0 ) . x y Proof. By the theorem above f ( z ) u ( x, y ) 1 v( x, y ) i has a complex derivative at z 0 x0 1 y 0 i if and only if u x ( x0 , y 0 ) v ( x , y ) x 0 0 u ( x0 , y 0 ) y v ( x0 , y 0 ) y r s P 1 P s r for some r and s . Since in this case P is 2 2 identity matrix, we obtain u x ( x0 , y 0 ) v ( x , y ) x 0 0 u ( x0 , y 0 ) y r s v s r ( x0 , y 0 ) y which yields the desired equalities. ■ The corollary above is known in literature as Cauchy-Riemann equations [ 3 ]. These equations are used almost as a definition of C - differentiability in textbooks. But notice that they are not basis free, whereas our definition is. References: [ 1 ] Howard Anton and Chris Rorres, Elementary Linear Algebra, John Wiley & Sons, Inc., 1994 [ 2 ] Walter Rudin, Principles of Mathematical Analysis, McGraw-Hill International Editions, 1976 [ 3 ] Serge Lang, Complex Analysis, New York: Springer, 1999