Eigenvalues for HP-41cv/cx

This program determines the eigenvalues of a real number matrix.

Theory

Multiplying a vector with a square matrix, usually changes both the magnitude (scale) and the direction of the vector.  In the special case where it changes only the magnitude of the vector and leaves the direction unchanged (or switches the vector to the opposite direction), that vector is called an eigenvector of that matrix.

When multiplied by a matrix, each eigenvector of that matrix changes its magnitude by a factor, called the eigenvalue corresponding to that eigenvector.  In other words: vector x is called an eigenvector of the matrix A with eigenvalue λ if the following equation holds [5]

A.x = λ.x

The eigenvalues of A are precisely the solutions λ to the equation det(A – λ.x) = 0. Here det() is the determinant of matrix formed by A – λ.I in which I is a n×n identity matrix. This equation is called the characteristic equation of A.

For a more detailed description including complex eigenvalues, refer to my article “Complex Eigenvalues for HP-41cv/cx”

Method

The program uses the iterative power method to determine the largest eigenvalue. This value is removed from the matrix using the deflation method of Hotelling [1]. The remaining eigenvalues are determined in a similar way. Note that accumulation of rounding errors will affect the accuracy of successive eigenvalues.

The power method is an iterative method for approximating eigenvalues. This method is perhaps not the most respected method, and can only be used when:

• There are no two identical eigenvalues.
• The start vector is not the null vector.
• The matrix does not contain complex eigenvalues. (The method will typically work well with a strongly dominated diagonal).

If desired, the eigenvectors can be calculated from the original matrix A. (The deflation method spoils the eigenvector, but not the eigenvalue.)

Examples

1. Matrix A
   

   keystrokes	display
   FIX 3
   XEQ “EW”	DIM?
   2 R/S	INV.MATRIX:
   	1:1=0.000?
   8 R/S	1:2=0.000?
   4 R/S	2:1=0.000?
   3 R/S	2:2=0.000?
   6 R/S	INV.STARTV.:
   	1:1=0.000?
   1 R/S	2:1=0.000?
   1 R/S	EPS=0.001?
   R/S	10.500
   	:
   	10.605
   	EW=10.606
   R/S	INV.STARTV.:
   	1:1=0.838?
   1 R/S	2:1=0.546?
   1 R/S	EPS=0.001?
   R/S	-0.230
   	3.393
   	3.394
   	EW=3.394
   R/S	EINDE

   The computed eigenvalues 10.606 and 3.394 can be verified through:
   

   This is also verified through the characteristic polynomial. The eigenvalues of A are precisely the solutions λ to the equation listed below with I is the n×n identify matrix.
   
   
   Therefore:
   
   The solutions to this equation are the eigenvalues λ_i.
   The JavaScript calculator “Eigenwerte und Eigenvektoren berechnen” provides another verification [7].

Listing

• Requires
  • X-Functions module on the HP-41cv
  • Advantage ROM
• Available as
  • source code
  • raw code for the V41 emulator
  • bar code for the HP Wand (HP82153A)
• Size
  • 39 registers
• Registers:
  • alfa, X, Y, Z, T, L
  • R00: epsilon
  • R01: loop counter
  • R02: eigenvalue
• Long jumps:
  • 41
  • 95

;  /---------------------------------------------------------------------\
;  |                      E i g e n v a l u e s                          |
;  |                                                                     |
;  |                            for the HP-41                            |
;  \---------------------------------------------------------------------/
;
;                                   1.00
;                               Coert Vonk
;
;            http://www.coertvonk.com/technology/hp41/eigenvalues-4606

01      LBL "EW"
02       E-4            ; default eps
03      STO 00
04      CF 08
05      "DIM ?"         ; invoer dimensie
06      PROMPT
07      ABS
08      INT
09      STO 01
10      ENTER^
11      ENTER^
12       E3
13      /
14      +
15      "INP. MATRIX:"
16      AVIEW
17      "B"
18      XROM 22,29      ; MATDIM: creeer B-matrix
19      "A"
20      XROM 22,29      ; MATDIM: creer A-matrix
21      XROM 22,61      ; MEDIT: invoer A-matrix

22      LBL 00
23      "A"
24      XROM 22,18      ; DIM?
25      INT
26      "INP. STARTV.:"
27      AVIEW
28      "W"             ; creer W-vector
29      XROM 22,29      ; MATDIM
30      "V"             ; creer V-vector
31      XROM 22,29      ; MATDIM
32      XROM 22,61      ; MEDIT: invoer V-vector
33      XROM 22,19      ; FNRM
34      "STARTV. ERR"
35      X=0?
36      AVIEW
37      X=0?
38      PSE             ; als startvector
39      X=0?            ; de nulvector is
40	GTO 00		; dan opnieuw
41	XEQ 03
42	RCL 00
43	"EPS="
44	ARCL X
45	>"?"
46	PROMPT		; invoer epsilon
47	STO 00
48	LBL 01
49	"W"
50	XROM 22,57      ; SUM
51	XROM 22,18      ; DIM?
52	INT
53	/
54	VIEW X		; tussenresultaat v/d e.g. laten zien
55	STO 02
56	RCL 00
57	XROM 22,33      ; MIN
58	+
59	XROM 22,30      ; MAX
60	X<=Y?		; test of er aan de nauwkeurigheid eps word voldaan
61	GTO 00		; zo  ja, spring
62	XEQ 03		; zo nee, ga verder
63	GTO 01

; toon de gevonden e.w.

64	LBL 00
65	RCL 02
66	"EV="
67	ARCL X
68	PROMPT
69	DSE 01		; alle ew nog niet gevonden?
70	GTO 04		; dan spring naar 04
71	"V"
72	PURFL		; maak de extended-memory weer schoon
73	"W"
74	PURFL
75	"A"
76	PURFL
77	"B"
78	PURFL
79	"EINDE"		; we're done
80	PROMPT
81	GTO E

; zeef de gevonden eigenwaarde uit de laatst gebruikte matrix (A)

82	LBL 04
83	 E
84	"X,V,W"
85	XROM 22,25      ; MAT*
86	"W"
87	XROM 22,59      ; TRNPS
88	"V,W,B"
89	XROM 22,24      ; M*M
90	RCL 02
91	"X,B"
92	XROM 22,25      ; MAT*
93	"A,B,A"
94	XROM 22,27      ; MAT-
95	GTO 00

; voer een iteratie slag uit volgens de powermethode
; zodat v:=A*v en w:=e.w. vector

96	LBL 03
97	"A,V,W"
98	XROM 22,24      ; M*M
99      "W,V"
100     XROM 22,28      ; MAT/
101     1.001
102     XROM 22,18      ; DIM?
103     RCL Y
104     "V,W"
105     XROM 22,49      ; MSWAP
106     "V"
107     XROM 22,19      ; FNRM
108     1/X
109     "X,V"
110     XROM 22,25      ; MAT*
111	RTN

112	END

References