Talk:Vigenère cipher/Cryptanalysis

I'm not completely sure how algorithmically feasible this task is, but it at least sounds interesting. (Well, I'm not sure how feasible it is to do this completely automatically.) I'll try to add an example soon. MagiMaster 05:37, 31 May 2011 (UTC)

C++ code added, but I'm not sure about how robust it is. There seems like a decent chance that the first part will pass a too-short key, and the second part doesn't have any checks on the final output. Another chi-squared check doesn't work, which also makes me suspect that the chi-squared check in the first part isn't the best solution either, even if it does ok. MagiMaster 11:28, 31 May 2011 (UTC)

There. I changed the chi-squared test to something that maximizes the correlation instead. Smaller pieces increase the correlation artificially, so I added a weight against them. It seems to work on nearly all the ciphertext I've tried. Long keys are more likely to cause errors, but it still gets the length and most of the characters. A second pass to try and correct any errors by looking for words or common trigrams might not be a bad idea, but I don't think I'll do that myself. MagiMaster 20:26, 31 May 2011 (UTC)

Task specification

I think that there should be a requirement to provide some output. Specifically each solution should output the key length determined, the discovered key, and some of the decrypted text (say a line). --Dgamey 14:14, 3 June 2011 (UTC)

Well, outputting the key gives the length of the key too, but yeah I could add some details to the output specification. MagiMaster 04:48, 4 June 2011 (UTC)

Difficulty and suggested improvements

This is a huge amount of text which makes the task much easier. As a hobby I used to solve these by hand and we would typically get cipher texts of 15-20 5 letter groups. This cipher text has 165 groups!

I suggest to make it more interesting, that we try our algorithms on successively shorter texts (halving the text each time) and see how low we can drive it.--Dgamey 12:28, 3 June 2011 (UTC)

Well, I put more text since this is supposed to be fully autonomous. I think if you can look at the output, look for word pieces, and use that to improve your guess, you wouldn't need as much text. That said, I did test the algorithm I posted on shorter inputs and it did fine. It was only when the key was fairly long compared to the text that it started to make mistakes. Also, the text I gave is a specific chunk of a certain work. I didn't put the output of my program on the page in case anyone wanted to view this as a puzzle. (Adding a second, shorter piece of ciphertext would be fine though.) MagiMaster 04:49, 4 June 2011 (UTC)