B Square Root of an Odd Powered Integer is Always Irrational?

[e2m2a]

TL;DR Summary

Excel giving strange results. Seems to fail for taking square roots of very large odd powered numbers.

Is it always true that the square root of an odd powered integer will always be irrational?

 

[PeroK]

Summary:: Excel giving strange results. Seems to fail for taking square roots of very large odd powered numbers.

Is it always true that the square root of an odd powered integer will always be irrational?

What about $\sqrt{4^3} = 8$?

 

[e2m2a]

Well, actually 4^3 can be expressed as (2^2)to the third power or (2^3)squared. What I meant is an integer that cannot be expressed as a square and then raised to an odd power. Would not such an integer raised to an odd power always yield an irrational number when the square root is taken?

 

[PeroK]

Well, actually 4^3 can be expressed as (2^2)to the third power or (2^3)squared. What I meant is an integer that cannot be expressed as a square and then raised to an odd power. Would not such an integer raised to an odd power always yield an irrational number when the square root is taken?

Take any positive integer $n$. Either:

a) $n$ is a perfect square, hence $\sqrt n$ is a postive integer.

or

b) $\sqrt n$ is irrational.

In other words, there is no positive integer $n$ where:

c) $\sqrt n$ is a proper rational. I.e. rational but not an integer.

Proving this is a useful exercise.

 

[e2m2a]

From the research I have done since posting, it says that the square root of any non-square integer is always an irrational number.

 

[mfb]

Correct, that's also what post 4 says. It's pretty easy to prove.

 

[pbuk]

Summary:: Excel giving strange results. Seems to fail for taking square roots of very large odd powered numbers.

Fail in what way?

Is it always true that the square root of an odd powered integer will always be irrational?

Whether something is irrational or not has absolutely no bearing on the accuracy of any compution in Excel.

 

[jbriggs444]

Whether something is irrational or not has absolutely no bearing on the accuracy of any compution in Excel.

Except, of course, to the extent that no irrational number can be a model number in IEEE 754 floating point. An integer might or might not be.

 

[e2m2a]

Fail in what way?Whether something is irrational or not has absolutely no bearing on the accuracy of any compution in Excel.

Excel rounds off the square root of a very large odd powered integer so that it gives an integer value. I have tested this with very large numbers. This is why I needed to research if the square root of odd powered integers always yields an irrational number. Excel like any other computer programs have a limitation on their accuracy.

 

[jbriggs444]

Excel rounds off the square root of a very large odd powered integer so that it gives an integer value. I have tested this with very large numbers. This is why I needed to research if the square root of odd powered integers always yields an irrational number. Excel like any other computer programs have a limitation on their accuracy.

Can you give an example? "Very large" is not very exact.

Edit: figured it out for you...

By default, Excel displays cells in "General" format. Using this format, you can put a formula into a cell and see a number displayed. For instance, =sqrt(3^41).

Under the default "General" format, this displays as 6039287738
Under the "Number" format, this displays as 6039287737.57
Under the "Accounting" format, this displays as 6,039,287,737.57
Under the "Number" format with 10 decimal places, this displays at 6039287737.5706100000

One assumes that those last five digits were simply zero filled. That is 15 or maybe 16 decimal digits of precision. Which is about right for a 64 bit IEEE float.

 

[pasmith]

Excel rounds off the square root of a very large odd powered integer so that it gives an integer value. I have tested this with very large numbers. This is why I needed to research if the square root of odd powered integers always yields an irrational number. Excel like any other computer programs have a limitation on their accuracy.

Excel might not do that internally, but the number of significant digits it displays might make it look like it does that.

 

[e2m2a]

Can you give an example? "Very large" is not very exact.

Edit: figured it out for you...

By default, Excel displays cells in "General" format. Using this format, you can put a formula into a cell and see a number displayed. For instance, =sqrt(3^41).

Under the default "General" format, this displays as 6039287738
Under the "Number" format, this displays as 6039287737.57
Under the "Accounting" format, this displays as 6,039,287,737.57
Under the "Number" format with 10 decimal places, this displays at 6039287737.5706100000

One assumes that those last five digits were simply zero filled. That is 15 or maybe 16 decimal digits of precision. Which is about right for a 64 bit IEEE float.

Raise 7 to the 39th power in excel and then take the square root. The answer should be an irrational number but it shows up as an integer with zeroes in the decimal place.

 

[PeroK]

Raise 7 to the 39th power in excel and then take the square root. The answer should be an irrational number but it shows up as an integer with zeroes in the decimal place.

When I try I get (same as using Python):

3.01586E+16​

 

[jbriggs444]

Raise 7 to the 39th power in excel and then take the square root. The answer should be an irrational number but it shows up as an integer with zeroes in the decimal place.

15 digits of precision displayed, followed by zeroes. The underlying implementation is 64 bit floating point. Apparently rounded to 15 decimal digits.

https://www.microsoft.com/en-us/mic...y-does-excel-give-me-seemingly-wrong-answers/

 

[Mark44]

15 digits of precision displayed, followed by zeroes. The underlying implementation is 64 bit floating point. Apparently rounded to 15 decimal digits.

The IEEE 754 Standard for floating point 64-bit numbers maintains slightly over 15 decimal digit precision. See https://en.wikipedia.org/wiki/IEEE_754.

 

[jbriggs444]

The IEEE 754 Standard for floating point 64-bit numbers maintains slightly over 15 decimal digit precision. See https://en.wikipedia.org/wiki/IEEE_754.

Right. But Excel seems to consistently chop them off at 15.

Try keying 1234567890123456 into a cell and set the display format to number (default 2 digits to the right of the decimal point). The six is discarded in favor of a zero. Same if you key in 1234567890.123456. [Excel actually truncates the cell contents in these cases, not just the displayed value]

Or, equally dramatic, try =111111111*111111111 (9 digits times 9 digits) for a result of 12345678987654300 (off by 21).

It is a reasonable strategy, producing zeroes rather than digits of questionable accuracy.

 

[mfb]

You can check if it has the next digit internally if you calculate 111111111*111111111 - 12345678987654200. Is the result 100 or something close to it?

 

[pasmith]

Raise 7 to the 39th power in excel and then take the square root. The answer should be an irrational number but it shows up as an integer with zeroes in the decimal place.

39\log7 \approx 32.96. I would not expect the result of 7^{39} to be accurate, still less the result of the square root.

 

[jbriggs444]

You can check if it has the next digit internally if you calculate 111111111*111111111 - 12345678987654200. Is the result 100 or something close to it?

Nicely designed experiment. The result is 120. Based on this result, the intermediate result used by Excel was 12345678987654320 exactly. This fits the IEEE 754 expectation.

Expressed in binary, the product 111111111*111111111 has 54 bits. IEEE 754 can express 53 (counting one extra "hidden" bit from normalization). Here is the true mathematical result for the product. [Result obtained from Windows calculator, programmer mode]

10 1011 1101 1100 0101 0100 0110 0010 1001 0001 1111 0100 1011 0001
 <---------------------------- 52 bits -------------------------->
^                                                                 ^
One "hidden" bit from normalization           One bit not expressed

IEEE 754 64 bit can exactly express the number one greater than this and the number one less than this. Those two are both "model numbers" within the floating point format. But the mathematically correct result is not a model number. IEEE 754 64 bit cannot express the product exactly.

The value that you asked to be subtracted (12345678987654200) is a model number which also fits in 15 decimal digits. (All 15 digit integers are model numbers within the 64 bit IEEE floats). Again, my compliments on the chosen probe.

One can get more information on IEEE 754 from the wiki page. I've extracted a summary here.

Digression alert...

Many programmers treat floating point as being inherently inexact. These individuals usually display a healthy paranoia in their programs. However, floating point has exact and predictable behavior.

Forty years ago I was using 64 bit floating point in an accounting environment for correct-to-the-penny calculations. [I hate bean counters]. Our hardware and programming languages provided no direct support for 64 bit integer arithmetic. Fortunately, the hardware and software did support 56 bit significands in its 64 bit (VAX D floating) format. Put an integer number of pennies into a D float and everything works fine as long as you are dealing with less than 720 trillion dollars or so.

 

[pbuk]

Excel like any other computer programs have a limitation on their accuracy.

Raise 7 to the 39th power in excel and then take the square root. The answer should be an irrational number but it shows up as an integer with zeroes in the decimal place.

But again this has nothing to do with rational vs irrational: Excel does not calculate sqrt(7^40) exactly either.

 

[sysprog]

You can check if it has the next digit internally if you calculate 111111111*111111111 - 12345678987654200. Is the result 100 or something close to it?

Windows 10 Calculator shows 111111111 x 111111111 = 12345678987654321 $-$ maybe the Excel guys should try to glean something from the Calculator code, now that it's open source (github.com/microsoft/calculator).

 

[pbuk]

Windows Calculator uses arbitrary precision arithmetic (some of the time) which requires a different set of choices for dealing with edge cases from arithmetic using numbers stored using IEEE 754.

The limitations of IEEE 754 are well suited to (and well understood by computational experts in) almost all applied scientific and financial purposes, which is what Excel is mainly used for.

If you want arbitrary precision, rational or Big Integer arithmetic then use a package that supports that.

 

[sysprog]

Windows Calculator uses arbitrary precision arithmetic (some of the time) which requires a different set of choices for dealing with edge cases from arithmetic using numbers stored using IEEE 754.

The limitations of IEEE 754 are well suited to (and well understood by computational experts in) almost all applied scientific and financial purposes, which is what Excel is mainly used for.

If you want arbitrary precision, rational or Big Integer arithmetic then use a package that supports that.

Yeah, when I increase the number of 1s to 17, the result is 1.2345679012345678765432098765432e+32.

 

[mfb]

and everything works fine as long as you are dealing with less than 720 trillion dollars or so.

Zimbabwe would like to have a word different implementation.

There are a few bank errors which can give some insight. This 11.7 quadrillion bill was simply adding a lot of zeros to the real bill (they offered a repayment plan!), but this $23 quadrillion dollar balance has a lot of different digits. No obvious pattern in decimal or binary dollars - but in cents, it's 10000000100000001000000010000000100000001000000001001001010000_2 or 2020202020201250_16. The pattern here is obvious. The last digits are caused by actual account transactions, and a computer error added a lot of 0x20 bytes to it. There are more than 53 non-zero binary digits and 62 in total. This suggests the bank uses 64 bit integers.

 

[Svein]

The last digits are caused by actual account transactions, and a computer error added a lot of 0x20 bytes to it.

ASCII 0x20 is " " (space). Don't blame the computer - as usual, the error can be traced to one or more inept programmers.

 

[mfb]

I'd like to deposit " " in my bank account please. That should be free, right?

 

[Nugatory]

However, floating point has exact and predictable behavior.

Yes, although that statement might easily be misunderstood by someone who hasn’t been through the wars. As many pre-IEEE floating point designs demonstrated, “exact and predictable” does not preclude “bizarre and surprising”.

IEEE-754 is one of the underappreciated triumphs of computing, making the world safer for the naive every day.

 

[BWV]

The nth root of x given by =int(x^(1/n)) in Excel is an integer ;)

 

[jbriggs444]

The nth root of x given by =int(x^(1/n)) in Excel is an integer ;)

Are you suggesting something like...

Put some positive integer into A1
Put some positive integer into A2
Put =A1^A2 into A3
Put =INT(A3^(1/A2)) into A4
Observe that A4 is equal to A1 unless an overflow has occurred while computing A3

 

[BWV]

Are you suggesting something like...

Put some positive integer into A1
Put some positive integer into A2
Put =A1^A2 into A3
Put =INT(A3^(1/A2)) into A4
Observe that A4 is equal to A1 unless an overflow has occurred while computing A3

It was just an attempt at a joke, but per the OP x would be a^some odd power and n would be 1/2

 

[Baluncore]

From the research I have done since posting, it says that the square root of any non-square integer is always an irrational number.

Did you notice that if you write any of those irrational square roots as a continued fraction, they always repeat.

 

[sysprog]

Is it always true that the square root of an odd powered integer will always be irrational?

What about 1?