# find solutions that solve the equations

I am implementing a timing attack on RSA for school and I need to generate two sets of messages $Y$ and $Z$ for which holds:

$(Y^d \mod N) \cdot Y < N$ and
$(Z^d \mod N) \cdot Z > N$

where $d$ and $N$ are known. How can I efficiently find solutions for $Y$ and $Z$?

I have tried using a random function but it takes too long to complete or it doesn't complete at all.

-
I agree with @CodesInChaos . If we assume $d$ is random, then $Y=k$ has probability $1/k$ of working; likely there is a small $Y$ that would work (and starting at the smallest and working your way up seems like the obvious approach (unless, of course, $Y=0$ or $Y=1$ isn't disallowed for some reason). –  poncho Sep 7 '12 at 19:57
Do you have a typo in the condition for $Z$? The $*Z>Z$ part is weird. –  CodesInChaos Sep 7 '12 at 19:57
For Z, try any Z > N/2; that should work. –  poncho Sep 7 '12 at 20:04
You state $d$ is known. By convention, $d$ stands for the private exponent (which is one of the targets if you're trying to recover the private key; if you get that, you've won). Do you really mean the private exponent $d$? Or, do you really mean the public exponent (and are using nonstandard terminology to describe it)? –  poncho Sep 7 '12 at 20:19
Oh, if you've recovered one of the CRT exponents (say, $d \bmod p$), then the rest is easy; see the answer in crypto.stackexchange.com/questions/1890/… –  poncho Sep 7 '12 at 20:26

Finding a $Z$ is easy. This condition is fulfilled for all $Z>n/2$, and most of the smaller values of $Z$ too. There are only a few values fulfilling the condition for $Y$, and since this is almost the negation of the condition for $Y$, most numbers will fulfill it.

When looking at the condition for $Y$, you can model $Y^d \mod N$ as a random number between 0 and $N$. This leads to a success chance of approximately $1/Y$.

Thus a good strategy for finding a $Y$ is starting with $Y = 2$ and incrementing by one on each attempt, i.e. trying 2, 3, 4,... This should find a number fulfilling $(Y^d \mod N) \cdot Y < N$ quite quickly.

Total chance of finding an $Y \le n$ is:

$1-\Pi^n_{i=2}(1-1/i) =$

$1 - \frac{(n-1)!}{n!} =$

$1- 1/n$

Which quickly converges to 1

-
sorry there was a typo in the Z condition. For the Y condition - what would be a good increase on each attempt (N is 512bit at least) –  blejzz Sep 7 '12 at 20:03
@blejzz The best increase would be 1. You simply want to try the smallest $Y$s first, since those have the highest success chance. –  CodesInChaos Sep 7 '12 at 20:10

I'll add one more point: if you are not able to find a $Y$ satisfying your condition using CodesInChaos's approach, here is one more approach you can try as a fallback.
Pick a small value $i$, set $Y=i^e \pmod{N}$, and try $Y$. Note that $Y^d \bmod N$ will be equal to $i$, and $Y$ can be modelled as a random number between 0 and $N$. This means that you have a success chance of approximately $1/i$, with this strategy.
So, you can try $i=2$, $i=3$, $i=3$, \dots, in succession until you find the first success. By the same argument CodesInChaos gives, there is likely to be a small $i$ for which this succeeds.
Again, this doesn't really add anything. There is no particular reason to prefer this strategy over CodesInChaos's, if both $e$ and $d$ are known. However, this is available as a fallback if CodesInChaos's method fails. Also, this method is available if $d$ is not known but $e$ is known.