Suganthi Velliah

Greenhorn

Posts: 17

posted 8 years ago

Hi All,

While calculating logarithm values in JDK1.4 it returns some unpredicted values. As Math.log() returns double so there is loss of precision and exact log base 10 values cannot be calculated.

For a CR in TP, a formula was given as:

DispCode = Int ( -1 * log10( abs(LsbRsln)))

which is written in Java as,

int dispcode = (int) (-1 * (Math.log(Math.abs(Double.parseDouble(lsbrsln))) / Math.log(10)));

For example, if lsbrsln = 0.001, the formula is expected to return 3, instead it returns 2. The following part of the formula

Math.log (Math.abs (Double.parseDouble(0.001))) / Math.log(10) is returning -2.999999� and finally the formula is returning 2 while casting it into int.

In JDK1.5 we have Math.log10 method itself for calculating logbase10 values. But since we need to support JDK1.4 also we were not able to use that also.

Please let me know if you have the solution.

While calculating logarithm values in JDK1.4 it returns some unpredicted values. As Math.log() returns double so there is loss of precision and exact log base 10 values cannot be calculated.

For a CR in TP, a formula was given as:

DispCode = Int ( -1 * log10( abs(LsbRsln)))

which is written in Java as,

int dispcode = (int) (-1 * (Math.log(Math.abs(Double.parseDouble(lsbrsln))) / Math.log(10)));

For example, if lsbrsln = 0.001, the formula is expected to return 3, instead it returns 2. The following part of the formula

Math.log (Math.abs (Double.parseDouble(0.001))) / Math.log(10) is returning -2.999999� and finally the formula is returning 2 while casting it into int.

In JDK1.5 we have Math.log10 method itself for calculating logbase10 values. But since we need to support JDK1.4 also we were not able to use that also.

Please let me know if you have the solution.

Ulf Dittmer

Rancher

Posts: 42970

73

posted 8 years ago

Using an "(int)" cast to create an integer from a floating point number is rather crude, especially since floating point arithmetic on computers is fundamentally imprecise (see #20 here to understand why). Depending on your circumstances, Math.round, Math.ceil or Math.floor may be a better way to do that.

Using the BigDecimal class may also help.

Using the BigDecimal class may also help.