Hi everyone,

I'm pretty new with sage. I was forced to change from MATLAB to Sage, because I was told Sage does approximate very tiny number better as it can work with sqrt(2) as sqrt(2) and not as the rational number approximating it.

Approximations are very important for my problem. I need to evaluate this integral (I write it Latex code, because due to my karma I cannot upgrade the picture with the integral)

```
\sum_{c=1}^{d}\int_{\min(256c-0.5,\frac{y(y+1)}{2})}^{\min(256c+0.5,\frac{y(y+1)}{2})}\frac{1}{\sqrt{2\pi}}e^{-\frac{x^2}{2}}\, dx
```

Suppose d = 1, then this is simply the integral

```
\int_{255.5}^{256.5}\frac{1}{\sqrt{2\pi}}e^{-\frac{x^2}{2}}\, dx
```

To evaluate this integral I wrote the following code

T = RealDistribution('gaussian,1) print T.cum_distribution_function(256.5)-T.cum_distribution_function(255.5)

because the integral above is the same as the difference of the distribution function of a standard distributed gaussian random variable between the boundaries of the integral. However, and you can check yourselves if you don't believe it, the result I get with sage is 0.

I guess that this is due to some approximation, which sage does. Indeed the exact value of the integral is pretty close (and with pretty I mean a lot) to 0. My problem is that I need to be able to have the exact value, because the sum of integrals I'm working with and my whole work behind this integral requires me to be very careful with those little tiny numbers.

I tried to use the function

integrate

to deal with this problem, but something funny and apparently inexplicable happened when I was trying to use it.

To be precise I defined this code:

```
def VarianceOfvy(y):
temp1 = 0
temp2 = 0
for r in range(0,y+1):
for x in range(0,r+1):
temp1 = temp1 + (255/256)^x * 1/256 * (r-x)^2
for r in range(0,y+1):
for x in range(0,r+1):
temp2 = temp2 + ((255/256)^x * 1/256 * (r-x))^2
sigma = temp1 - temp2
return sqrt(sigma)
def Integerd(y):
b = y*(y+1)/2
d = 1
c = 0
while min((c+1)*256-0.5,b) == (c+1)*256-0.5:
d = d+1
c = c+1
return d
def Probabilityvynequiv(y):
var('c')
b = (y*(y+1))/2
sigma = 2
mu = 1
d = Integerd(y)
factor = 1/(integrate(1/sqrt(2*pi)*e^(-(x/sigma)^2/2),x,-oo,(b-mu)/sigma) - integrate(1/sqrt(2*pi)*e^(-(x)^2/2),x,-oo,(-mu)/sigma))
p = sum(factor*1/sigma*integrate(1/sqrt(2*pi)*e^((-x^2)/(2)),x,c*256+0.5,min((c+1)*256-0.5,b)),c,0,d)
return p
```

And if I let it run, the result I get is

```
1/2*(erf(255.75*sqrt(2)) - erf(128.25*sqrt(2)) + erf(127.75*sqrt(2)) - erf(0.25*sqrt(2)))/(3*erf(1/4*sqrt(2)) + 1)
```

which I assume is correct, and at least it tells me that Sage is able to read my code and output a result. If I call the function VarianceOfvy(2), the result I get is 3/65536*sqrt(11133895), which is also correct. Now, if I'm changing the command

```
sigma = 2
```

with

```
sigma = VarianceOfvy(2)
```

and try to let the whole program run again, Sage is not able anymore to output a result.

I'm really lost and I don't know what to do. Could someone advise me and give me some hints on how to evaluate those tiny integrals, in such a way that I don't loose any approximation?