ArithmeticError on big numbers?

So I was just writing an answer to this question and then I checked

given:

``````defmodule Pow do
def powa(x, n), do: powa(x, n, 1)
def powa(_, 0, acc), do: acc
def powa(x, n, acc), do: powa(x, n-1, acc * x)
end
``````

`Pow.powa(phi, 1000)` gives 3.2735165627676025e209

But

``````iex(65)> Pow.powa(phi, 10000)
** (ArithmeticError) bad argument in arithmetic expression: 1.5724845684981474e308 * 1.62
``````

What? Why? And How do we fix this?

You have exceeded the size that can be fit into a 64 bit float. There is no fixing this without switching to integers.

3 Likes

This is a limitation in floating point arithmetic, the value gets to big, to be storable in 64 bit floating point.

1 Like

interestingly

`Pow.powa(2, 10000)` was working

`19950631168807583848837421626835850838234968318861924548520089498529438830221946631919961684036194597899331129423209124271556491349413781117593785932096323957855730046793794526765246551266059895520550086918193311542508608460618104685509074866089624888090489894838009253941633257850621568309473902556912388065225096643874441046759871626985453222868538161694315775629640762836880760732228535091641476183956381458969463899410840960536267821064621427333394036525565649530603142680234969400335934316651459297773279665775606172582031407994198179607378245683762280037302885487251900834464581454650557929601414833921615734588139257095379769119277800826957735674444123062018757836325502728323789270710373802866393031428133241401624195671690574061419654342324638801248856147305207431992259611796250130992860241708340807605932320161268492288496255841312844061536738951487114256315111089745514203313820202931640957596464756010405845841566072044962867016515061920631004186422275908670900574606417856951911456055068251250406007519842261898059237118054444788072906395242548339221982707404473162376760846613033778706039803413197133493654622700563169937455508241780972810983291314403571877524768509857276937926433221599399876886660808368837838027643282775172273657572744784112294389733810861607423253291974813120197604178281965697475898164531258434135959862784130128185406283476649088690521047580882615823961985770122407044330583075869039319604603404973156583208672105913300903752823415539745394397715257455290510212310947321610753474825740775273986348298498340756937955646638621874569499279016572103701364433135817214311791398222983845847334440270964182851005072927748364550578634501100852987812389473928699540834346158807043959118985815145779177143619698728131459483783202081474982171858011389071228250905826817436220577475921417653715687725614904582904992461028630081535583308130101987675856234343538955409175623400844887526162643568648833519463720377293240094456246923254350400678027273837755376406726898636241037491410966718557050759098100246789880178271925953381282421954028302759408448955014676668389697996886241636313376393903373455801407636741877711055384225739499110186468219696581651485130494222369947714763069155468217682876200362777257723781365331611196811280792669481887201298643660768551639860534602297871557517947385246369446923087894265948217008051120322365496288169035739121368338393591756418733850510970271613915439590991598154654417336311656936031122249937969999226781732358023111862644575299135758175008199839236284615249881088960232244362173771618086357015468484058622329792853875623486556440536962622018963571028812361567512543338303270029097668650568557157505516727518899194129711337690149916181315171544007728650573189557450920330185304847113818315407324053319038462084036421763703911550639789000742853672196280903477974533320468368795868580237952218629120080742819551317948157624448298518461509704888027274721574688131594750409732115080498190455803416826949787141316063210686391511681774304792596709376`

so I thought it could be some other issue.

when you say switching to integers, in this case it is not possible right?

Thats integers.

If you want to have `phi` which is some irrational constant as far as I remember. So you canâ€™t make that an integer. You could though â€śroundâ€ť at some level.

1 Like

just not happening with
1.5724845684981474e308 * 1.6

Thatâ€™s what heâ€™s saying. `1.5724845684981474e308 * 1.6` is a float, not an integer. `2` is an integer, not a float. Floats have a 64 bit size limit. Integers in Erlang are limited only by the available ram.

3 Likes

If you are willing to understand the issue deeper, Wikipediaâ€™s article on IEEE-754 is quite good.

1 Like