Practical and Efficient Evaluation of Inverse Functions

Transcription

1 J. C. HAYEN ORMATYC 017 TEXT PAGE A-1 Practical and Efficient Evaluation of Inverse Functions Jeffrey C. Hayen Oregon Institute of Technology ORMATYC 017

2 J. C. HAYEN ORMATYC 017 TEXT PAGE A- PRELIMINARY EXERCISE Evaluation of a Recurrence Relation (for a Mystery Function) k1 k( bk) Based upon the relation above, with b 7 and 0 0., complete the table below: b k k k 1 g( b) Epress the results entered (in columns 3 and 4) to 5 decimal places of precision.

3 J. C. HAYEN ORMATYC 017 TEXT PAGE A-3 A SET-THEORETIC APPROACH TO RELATIONS AND FUNCTIONS Definition (A Cartesian Product) Let A and B be sets. The set of all ordered pairs, with the first element in A and the last element in B, is called the Cartesian product of A with B, which is denoted by A B. Epressed in set-builder notation, Definition (A Relation) A B{ (, y): A, y B} Let A and B be sets. A relation from A to B is a subset of A B. Let R denote this relation, whereby R A B. Suppose that (, y) R. Then this association between and y may be conveniently epressed as Ry. Definition (The Domain and Range of a Relation) The domain and range of a relation R A B, respectively denoted by Dom( R ) and Rng( R ), are the sets defined by

4 J. C. HAYEN ORMATYC 017 TEXT PAGE A-4 { } { } Dom( R) A: ybwith Ry, Rng( R) yb: Awith Ry Definition (A Function) A function from A to B is a relation from A to B which has the essential properties identified below. Let f denote this relation, and suppose that ( ab, ) association between a and b is customarily epressed as f ( a) f. Then this b. Properties of f : 1. Dom( f ) A. ( ab, ) fand( ac, ) f c b Also, the aggregate of f along with its domain A and co-domain B is often denoted by f : A B, and ( ab, ) is routinely called an input-output pair (associated by f ). Definition (A 1-to-1 Function) A function f : A B is said to be one-to-one (or, equivalently, 1-to-1) if, and only if, ( ab, ) fand( db, ) f d a Alternatively epressed, f ( d) f( a) d a.

5 J. C. HAYEN ORMATYC 017 TEXT PAGE A-5 Definition (An Onto Function) A function f : A B is said to be onto if, and only if, Rng( f ) B. Definition (A 1-to-1 Correspondence) A function f : A B is said to be a one-to-one correspondence if, and only if, 1. f is 1-to-1. f is onto Theorem (An Inverse Function) Suppose that f : A B is a 1-to-1 correspondence. Then y B, A (which is unique) such that f ( ) y. This property induces a function g: B g( y) f( ) y A for which g is called the inverse function for f, and it is a 1-to-1 correspondence as well. 1 Remark: g may be denoted by f 1 ; alternatively, f may be denoted by g. Thus, g or f may be regarded as the inverse of (or the original for) f or g, respectively.

6 J. C. HAYEN ORMATYC 017 TEXT PAGE A-6 INTERPRETATIONS AND PROPERTIES OF 1-TO-1 CORRESPONDENCES A Mapping Diagram for an Invertible Function (Figure 1) Emphasis: Association of each input with a unique output. A Process Diagram for an Invertible Function (Figure ) Emphasis: Conversion of each input into a unique output. Remark: y f( ) and g( y) yield the same graphs. Remark: y f( ) and y g ( ) yield different graphs. An Epanded Diagram based upon f as the Original Function (Figure 3) Cancellation Property 1: g( f( )) An Epanded Diagram based upon g as the Original Function (Figure 4) Cancellation Property : f ( g( y)) y

7 J. C. HAYEN ORMATYC 017 TEXT PAGE A-7 INPUT-OUTPUT TABLES FOR ORIGINAL/INVERSE FUNCTIONS Eample: f ( ) {} and g( y) y with A 0 and B 0 {} Table 1. Output f (Input) Input Output Table. Output g(input) Input Output [ ] Issue Arbitrary Inputs for instance: g(3)? 1 g(3) g(3) 1.73

8 J. C. HAYEN ORMATYC 017 TEXT PAGE A-8 INPUT-OUTPUT TABLES FOR ORIGINAL/INVERSE FUNCTIONS Eample: f( ) and g( y) log ( y) with A and B Table 3. Output f (Input) Input Output Table 4. Output g(input) Input Output [ ] Issue Arbitrary Inputs for instance: g(3)? 1 g(3) g(3) 1.585

9 J. C. HAYEN ORMATYC 017 TEXT PAGE A-9 ANALYTICAL DETERMINATION OF INVERSE FUNCTIONS Eample: Inverse Hyperbolic Cosine Function f( ) cosh( ) e e ; 0 1 ( ) 1 ( ) y e e e e y0 ( e ) y( e ) 10 g( y) 1 g( y) cosh ( y) ln y y 1 ; y1 Eample: Inverse Hyperbolic Sine Function f( ) sinh( ) e e ; 1 ( ) 1 g( y) sinh ( y) ln y y 1 ; y

10 J. C. HAYEN ORMATYC 017 TEXT PAGE A-10 STANDARD NEWTON-RAPHSON METHOD Introduction to the Method The Newton-Raphson method is a numerical algorithm utilized to obtain an accurate but approimate solution or root to an equation of the specific form F( ) 0 (1) where F( ) represents any epression that involves a single variable, presumed to be denoted by in this instance. In advanced courses on numerical analysis, it can be shown that this method (a) has desirable convergence properties, and (b) yields a desired root to Eq. (1) of suitable precision if such a solution eists. The algorithm inherently employs an iterative process. Typically, the solve utility or equivalent feature available on scientific calculators is based upon a practical implementation of the Newton-Raphson method, which is commonly known as the Secant method.

11 J. C. HAYEN ORMATYC 017 TEXT PAGE A-11 An important caveat of both these methods is that a sufficiently-close initial guess for the solution of interest must be provided in order for the iterative process to be successful. However, techniques to obtain an appropriate initial guess are available. Derivation of the Algorithm Essential Concept: Locally approimate the function F( ) with a linear function in order to forecast a desired root p for which F( p) 0. Accordingly, consider y b( ) c, y F( ) () a where is assumed to be close to p. This linear function should serve to fulfill its intended purpose if the conditions stipulated below are satisfied: e y a dya d e dye (3) After application of these conditions, it is determined that

12 J. C. HAYEN ORMATYC 017 TEXT PAGE A-1 As a result, the relation for y a in Eqs. () becomes b F( ), c F( ) (4) ya F( )( ) F( ) (5) Net, suppose that p is a root to Eq. (1), and that is nearby p, so is regarded as an estimate for p. Then ye 0 since F( p) 0. But y a is presumed to acceptably approimate y e. As a result, let k and k 1 for which ya 0, so that Eq. (5) yields the relation 0 F( )( ) F( ) (6) k k1 k k It is desirable and effective to convert this relation into a mechanism to generate an increasingly accurate sequence of values that estimate the root p: k1 k F( k ) F ( ) k (7)

13 J. C. HAYEN ORMATYC 017 TEXT PAGE A-13 where k is a previous estimate for p while k 1 is an updated estimate for p. This relation is recursively applied to refine and improve each estimate for p, and it can be proven that the sequence of estimates { k } rapidly converges to the desired p under modest conditions on the function F( ) and the initial guess 0. In practice, the iterative process (and the sequence itself) is terminated when a specified degree of precision in the accuracy of the most recent estimate for p is achieved. See Figure 5 for an insightful depiction of the iterative process and its effectiveness. EVALUATION OF INVERSE FUNCTIONS (VIA ORIGINAL FUNCTIONS) In applied sciences, it is often convenient to epress certain formulas via an inverse function g that is based upon an original function f. It then becomes necessary to evaluate g at a particular input b. Let denote the output of g in this instance: g( b) (8) From the unique-output property of functions along with the cancellation properties

14 J. C. HAYEN ORMATYC 017 TEXT PAGE A-14 of invertible functions, Eq. (8) can be recast in terms of the original function f as f ( ) f( g( b)) f( ) b (9) In order to utilize the method of Eq. (7), Eq. (9) must be realigned with Eq. (1): Eample Evaluate b for (a) b 1 3, and (b) b 7. F( ) 0 ; F( ) f( ) b (10) F( ) b b b F ( ) Based upon Eq. (7), the algorithm for the evaluation of b becomes b 1 ( ) k1 k k Graphical Estimates and Spreadsheet Results

15 J. C. HAYEN ORMATYC 017 TEXT PAGE A-15 b 0.5 b b7.5 b Eample Evaluate ln ( b ) for (a) b 1 3, and (b) b 7. F( ) e b ln ( b) e b F ( ) e Based upon Eq. (7), the algorithm for the evaluation of ln ( b ) becomes k 1 k 1 k be Graphical Estimates and Spreadsheet Results b 1.0 ln ( b) b 7.0 ln ( b)

16 J. C. HAYEN ORMATYC 017 TEXT PAGE A-16 Eample Evaluate 1 sin ( b) for (a) 4 6 b 5, and (b) b. 5 1 F( ) sin( ) b sin ( b) sin ( ) b F ( ) cos( ) Based upon Eq. (7), the algorithm for the evaluation of 1 sin ( b) becomes k1 k tan ( k) bsec( k) Graphical Estimates and Spreadsheet Results b 1.0 sin ( b) b 1.0 sin ( b) is undefined EXTENDED NEWTON-RAPHSON METHOD Eplanation of the Method

17 J. C. HAYEN ORMATYC 017 TEXT PAGE A-17 This method is developed in almost the same manner as in the case of the standard method, ecept that the relations indicated below are utilized in an effort to forecast a desired root p for which F( p) 0: a y a( ) b( ) c, y F( ) (11) where is assumed to be close to p. It is anticipated that the quadratic function will fulfill its intended purpose even better than the linear function employed for the standard method if the conditions stipulated below are satisfied: e y a dya d e a d ye d y d dye (1) After application of these conditions, it is determined that

18 J. C. HAYEN ORMATYC 017 TEXT PAGE A-18 Presentation of the Algorithm a 1 F( ), b F( ), c F( ) (13) By proceeding with the same approach as in the case of the standard method, again with and 1 for which ya 0, the relation for y a in Eqs. (11) yields k k k1 b b 4ac k (14) a where a, b, and c are evaluated as indicated in Eqs. (13), ecept that is replaced by k. At this stage, for reasons to be discussed, it is advantageous to rationalize the numerator of Eq. (14), which then becomes k1 k c b b 4ac (15) It is now apparent that an ambiguity eists for the selection of the proper sign so as

19 J. C. HAYEN ORMATYC 017 TEXT PAGE A-19 to develop a useful recurrence relation for the algorithm of interest. This ambiguity can be resolved by first performing a simplification: k1 k c b b 14 ac/ b (16) Net, by carefully eamining representative cases (see diagrams in Figure 6), it can be concluded that the proper sign to select is sgn ( b ). As a result, k1 k c bsgn ( b) b 14 ac/ b (17) or, since sgn ( b) b b for any real number b, k1 k / cb / 1 1 4ac b (18) Last, substituting the actual epressions for a, b, and c into Eq. (18) reveals

20 J. C. HAYEN ORMATYC 017 TEXT PAGE A-0 F( k ) k1 k ( k) F( ) k (19) for which ( k ) acts as a modulating factor that accelerates convergence, where ( k), ( k) F( k) F( k) F( ) 1 1 ( ) [ ] k k (0) These particular functions possess some important properties: 1. ( k ) 0 and ( k ) 1 as F( k ) 0. ( k ) 0 and ( k ) 1 if F( k ) 0 EVALUATION OF INVERSE FUNCTIONS (VIA ORIGINAL FUNCTIONS) The previous eamples are now reconsidered via the etended method. (See Progression of Calculations in Spreadsheet Results)

21 J. C. HAYEN ORMATYC 017 TEXT PAGE A-1 IMPROVED NEWTON-RAPHSON METHOD Outline of the Method This method is still based upon the etended method presented above. This method utilizes a different technique to estimate the initial guess. The initial guess is bracketed via supplying reasonable upper and lower bounds on the value of for which f ( ) b. This procedure is demonstrated in a spreadsheet to be presented. An important element of this improved method is the formation of a locally interpolating quadratic polynomial for f ( ): a y a bc, y f( ) (1) where y a is assumed to closely follow y e between these bounds on. But the most important characteristic of this quadratic polynomial is that, for a particular y a, it can be solved for the required. Thus, in the case of ya be selected as this solution for. e b, the initial guess 0 will

22 J. C. HAYEN ORMATYC 017 TEXT PAGE A- Procedure for the Method Two bounds on must be supplied: LB and UB, with LB UB, such that f( ) b f( ) if f is strictly increasing LB LB UB UB or f( ) b f( ) if f is strictly decreasing () Such bounds eist because the function f must be strictly monotonic in the vicinity of the value of for which f ( ) b, a necessary condition for invertible functions. Based upon these bounds, let y f( ) 1 LB MP LB UB ( ), y f( ) y f( ) 3 UB 3 3 (3) The interpolating quadratic polynomial is then determined by the conditions

23 J. C. HAYEN ORMATYC 017 TEXT PAGE A a b c y a b c y a b c y (4) By means of the rule of Cramer, this system of linear equations can be solved for the variables ( a, b, c ): a b c y1( 3 ) y( 13) y3( 1) ( )( )( ) y ( ) y ( ) y ( ) ( )( )( ) y13( 3) y 31( 13) y31( 1) ( )( )( ) (5) Finally, the initial guess 0 is determined from the proper solution to the equation a bc b (6)

24 J. C. HAYEN ORMATYC 017 TEXT PAGE A-4 whose well-known candidate solutions are, b b 4 a( c b) (7) a The sign ambiguity apparent above is resolved as indicated below: if 0 LB UB or if 0 LB UB (8) With 0 determined, the etended method is then applied eactly as before, by means of Eq. (19), to complete the implementation of the improved method. EVALUATION OF INVERSE FUNCTIONS (VIA ORIGINAL FUNCTIONS) The previous eamples are now reconsidered via the improved method. (See Progression of Calculations in Spreadsheet Results)

25

26

27

28

29 J. C. Hayen ORMATYC 017 Graphics Page C-1 Reciprocal Function Standard Method b k k k +1 = g (b )

30 J. C. Hayen ORMATYC 017 Graphics Page C- Square-Root Function Standard Method b k k k +1 = g (b ) Natural Logarithm Function Standard Method b k k k +1 = g (b ) Inverse-Sine Function Standard Method b k k k +1 = g (b )

31 J. C. Hayen ORMATYC 017 Graphics Page C Graphical Estimation of Square-Root Function y

32 J. C. Hayen ORMATYC 017 Graphics Page C Graphical Estimation of Natural Logarithm Function y

33 J. C. Hayen ORMATYC 017 Graphics Page C-5.50 Graphical Estimation of Inverse-Sine Function y

34 J. C. Hayen ORMATYC 017 Graphics Page C-6 Square-Root Function Etended Method b k k F ( k ) F' ( k ) F" ( k ) k +1 = g (b ) Natural Logarithm Function Etended Method b k k F ( k ) F' ( k ) F" ( k ) k +1 = g (b ) Inverse-Sine Function Etended Method b k k F ( k ) F' ( k ) F" ( k ) k +1 = g (b )

35 J. C. Hayen ORMATYC 017 Graphics Page C-7 Natural Logarithm Function Improved Method b k k F ( k ) F' ( k ) F" ( k ) k +1 = g (b ) LB MP UB f ( LB ) f ( MP ) f ( UB ) Inverse-Sine Function Improved Method b k k F ( k ) F' ( k ) F" ( k ) k +1 = g (b ) LB MP UB f ( LB ) f ( MP ) f ( UB ) Inverse-Tangent Function Improved Method b k k F ( k ) F' ( k ) F" ( k ) k +1 = g (b ) LB MP UB f ( LB ) f ( MP ) f ( UB )

36 ENGINEERING Dynamics II (MECH 31) Eplicit Time Solutions An Eample Problem Vertical Projectile Motion with a Drag Force Vertical Projectile Motion Formulas The formulas given below eactly describe the motion of a projectile which may be modeled as a particle, has been vertically projected in a uniform gravitational field, and is subjected to a drag force that is proportional to the projectile speed squared: F kv Note: In the formulas below, k, v o, m, and g o should be regarded as known parameters. Ascent Relations: 0 t ta { [ ]} m y ln 1 kvo/ ( mgo) cos kgo/ m ( ta t) k / [ / ] v mg k tan kg m ( t t) o o a d / / [ ] 1 a o o o t m ( kg ) tan kv ( mg ), m h ln 1 kvo/ ( mgo) k [ ] Descent Relations: ta t t i { [ ]} m y ln 1 kvo/ ( mgo) sech kgo/ m ( t ta) k / [ / ] v mg k tanh kg m ( t t ) o o a / / [ ] 1 d o o o t m ( kg ) sinh kv ( mg ), v i v o o / 1 kv ( mg ) o Mathematical Definitions sech ( u) e e tanh ( u) e e [ ] 1 sinh ( u) ln u 1u Note: All of the relations above can be simplified via a parameter, which is defined below. u u u e e u u u

37 ENGINEERING Notation/Terminology t : elapsed time (from the projection instant) y : projectile position (vertical displacement) v : projectile speed k : drag coefficient v o : projection speed (initial speed) m : particle mass g o : gravity constant : medium resistance intensity, defined by o kv /( mg ) o t a : ascent time h : maimum height t d : descent time v i : impact speed t i : impact instant, defined by t i t a t d Note: t a and t d denote the durations of the ascent and descent phases, respectively, whereas t i denotes the terminal value of t or (equivalently) the duration of the entire projectile motion.

38