The Museum of HP Calculators

HP Articles Forum

[Return to the Index ]
[ Previous | Next ]


Hyperbolic Approximations for Trigonometric Functions on the HP-12C

Posted by Gerson Washiski Barbosa on 26 Feb 2005, 1:27 p.m.

----------------------------------------------------------------

COS/SIN: 01 ENTER ACOS: 52 ENTER 02 ENTER 53 * 03 ENTER 54 1/x 04 68.3393 55 1 11 / 56 - 12 g e^x 57 g SQRT 13 ENTER ATAN: 58 1 14 1/x 59 x<>y 15 + 60 g x=<y 16 CHS 61 g GTO 65 17 4 62 1/x 18 + 63 CHS 19 x<>y 64 9 20 ENTER 65 0 21 * 66 x<>y 22 21813 67 g e^x 27 / 68 ENTER 28 g e^x 69 1/x 29 2 70 - 30 * 71 g LSTx 31 / 72 ENTER 32 ROLLDN 73 1/x 33 * 74 + 34 g SQRT 75 g LSTx 35 / 76 g LN 36 x<>y 77 ROLLDN 37 ENTER 78 / 38 * 79 1844.6 39 CHS 85 * 40 1 86 ROLLDN 41 + 87 ROLLDN 42 g SQRT 88 ROLLDN 43 * 89 6 44 x<>y 90 y^x 45 g GTO 00 91 g SQRT ASIN: 46 ENTER 92 CHS 47 * 93 32.2 48 CHS 97 + 49 1 98 / 50 + 99 + 51 g SQRT

----------------------------------------------------------------

Usage:

Trigonometric functions:

Enter angles in degrees, -90 =< x <= 90 (*):

R/S => cos(x) x<>y => sin(x) x<>y / => tan(x)

Or,

R/S / => tan(x)

(*) (actual limits for signs consistency: +/- 89.99998084)

Average percent error: 0.05% (0 to 89.98 degrees).

Inverse trigonometric functions:

Enter arguments within the following ranges:

0 =< x < 1 for arcsin(x); 1E-50 < x <= 1 for arccos(x); 0 =< x < 1E100 for arctan(x) (**)

GTO.46 R/S => arcsin(x) GTO.52 R/S => arccos(x) GTO.58 R/S => arctan(x)

Notice that the numeric keys cross-pattern below makes remembering the entry points addresses much easier:

|8|

|4| |5| |6|

|2|

central key + upper key: arctan(x) central key + lower key: arccos(x) left key + right key: arcsin(x)

(**) (though ARCTAN works for negative arguments up to -1, this should not be taken into account in order to keep all functions in the first quadrant.

Average ARCTAN percent error: 0.02% within the valid range (maximum 0.06% at x=1)

Some examples:

sin cos tan

0.0001 0.0000 1.0000 0.0000 (0.0000) (1.0000) (0.0000) (exact value at this precision)

-1.0000 -0.0175 0.9998 -0.0175 (-0.0175) (0.9994) (-0.0175)

15.0000 0.2593 0.9658 0.2685 (0.2588) (0.9659) (0.2679)

45.0000 0.7077 0.7065 1.0016 (0.7071) (0.7071) (1.0000)

89.0000 0.9998 0.0175 57.2041 (0.9998) (0.0175) (57.2900)

89.9900 1.0000 1.7450E-4 5730.7974 (1.0000) (1.7453E-4) 5729.5779

89.9999 1.0000 1.4120E-6 708193.48 (1.0000) (1.7453E-6) 572957.80

asin acos atan

0.0000 0.0000 Error 0 0.0000 (0.0000) (90.0000) (0.0000)

0.0001 0.0057 89.9943 0.0057 (0.0057) (89.9943) (0.0057)

0.1000 5.7383 84.2617 5.7097 (5.7392) (84.2608) (5.7106)

0.9999 89.1899 0.8101 45.0239 (89.1897) (0.8103) (44.9971

1.0000 Error 0 0.0000 45.0268 (90.0000) (0.0000) (45.0000)

2.0000 - - 63.4241 (63.4349)

5.0000 - - 78.6904 (78.6901)

999.9999 - - 89.9427 (89.9427)

1E6 - - 89.9999 89.9999

1E7 - - 90.0000 90.0000

Running times are about 4 to 5.5 seconds, depending on the function (not argument dependable).

--------------------------------------------------------------

Although there are a few excellent programs for computing trigonometric functions in the HP-12C, I decided to write this one to fit some specific needs: reasonable running time, arguments and results in degrees and accurate enough for any practical purposes. The running times are not as good as I initially thought they would be by using this approach, but are reasonable. As of the accuracy, it is comparable to a good slide ruler: about 0.05%. This program uses all 99 memory steps but leaves all remaining registers free (RO to R6 and the financial registers), however the stack is not preserved. The expressions (I) and (II) below are a fair approximation to cosine and arctangent. The dividends are hyperbolic functions close to their respective trigonometric functions and agreeing exactly at the extremes of their valid ranges. The divisors were obtained through curve fitting.

(I) cos(x) ~ (2-cosh(0.83840144x))/exp(0.15049479x^2), x in radians ( 0.83840144 = 2 acosh(2)/pi )

(-pi/2 =< x =< pi/2)

or, for programming optimization:

cos(x) ~ (4-(exp(x/68.3393)+1/(exp(x/68.3393))))/(2exp(x^2/21813)), x in degrees,

and

(II) arctan(x) ~ 1.03125550*tanh(x)/(1.03143767-0.032032226*x^3), results in radians (1.03125550 = pi/(4*tanh(1)),

(0 =< x =< 1)

or

arctan(x) ~ 1.03125550((exp(x)-1/exp(x))/(exp(x)+1/exp(x)))/(1.03143767-0.032032226x^3), results in radians

or, for programming optimization:

arctan(x) ~ 1844.6((exp(x)-1/exp(x))/(exp(x)+1/exp(x)))/(32.2-(abs(x))^3); (0 =< x < 1) and

arctan(x) ~ 90+1844.6((exp(-1/x)-1/exp(-1/x))/(exp(-1/x)+1/exp(-1/x)))/(32.2-(abs(-1/x))^3); x>=1, results in degrees,

The other functions are computed as follows:

sin(x)=sgn(x)*sqrt(1-cos^2) (Since sgn(x) is computed as sqrt(x^2)/x, sin(x) fails for x=0)

tan(x)=sin(x)/cos(x)

arcsin(x)=arccos(sqrt(1-x^2)) (This does not preserve the sign of the function, but allows it to be

computed sequentially, as an argument to ARCCOS)

arccos(x)=arctan(sqrt(1/x^2-1)) (This causes the loss of the sign, but requires less programming steps).

Plots of (I) and (II) may show to what extent these approximations are reliable.

Edited: 7 July 2005, 9:49 p.m.

Password:

[ Return to the Message Index ]

Go back to the main exhibit hall