22.09.2013 Views

SSL II USER'S GUIDE - Lahey Computer Systems

SSL II USER'S GUIDE - Lahey Computer Systems

SSL II USER'S GUIDE - Lahey Computer Systems

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

2.9 UNIT ROUND OFF<br />

<strong>SSL</strong> <strong>II</strong> subroutines frequently use the unit round off.<br />

The unit round off is a basic concept in error analysis<br />

for floating point arithmetic.<br />

Definition<br />

The unit round off of floating-point arithmetic are<br />

defined as follows:<br />

u = M 1-L / 2, for (correctly) rounded arithmetic<br />

u = M 1-L , for chopped arithmetic,<br />

where M is the base for a floating-point number system,<br />

and L is the number of digits used to hold the mantissa.<br />

In <strong>SSL</strong> <strong>II</strong>, the unit round off is used for convergence<br />

criterion or testing loss of significant figures.<br />

Error analysis for floating point arithmetic is covered in<br />

the following references:<br />

[1] Yamashita, S.<br />

On the Error Estimation in Floating-point<br />

Arithmetic<br />

Information Processing in Japan Vol. 15, PP.935-<br />

939, 1974<br />

[2] Wilkinson, J.H.<br />

Rounding Errors in Algebraic Process<br />

Her Britannic Majesty’s Stationery Office,<br />

London 1963<br />

GENERAL RULES<br />

2.10 ACCUMULATION OF SUMS<br />

Accumulation of sums is often used in numerical<br />

calculations. For instance, it occurs in solving a system<br />

of linear equations as sum of products, and in<br />

calculations of various vector operations.<br />

On the theory of error analysis for floating point<br />

arithmetic, in order to preserve the significant figures<br />

during the operation, it is important that accumulation of<br />

sums must be computed as exactly as possible.<br />

As a rule, in <strong>SSL</strong> <strong>II</strong> the higher precision accumulation is<br />

used to reduce the effect of round off errors.<br />

2.11 <strong>Computer</strong> Constants<br />

This manual uses symbols to express computer hardware<br />

constants. The symbols are defined as follows:<br />

• flmax: Positive maximum value for the floating-point<br />

number system<br />

(See AFMAX in Appendix A.)<br />

• flmin: Positive minimum value for the floating-point<br />

number system<br />

(See AFMIN in Appendix A.)<br />

• tmax: Upper limit of an argument for a trigonometric<br />

function (sin and cos)<br />

Upper limit of argument Application<br />

Single<br />

precision<br />

8.23 x 10 5 FACOM M series<br />

FACOM S series<br />

Double<br />

precision<br />

3.53 × 10 15 SX/G 100/200 series<br />

FM series<br />

27

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!