Take the 2-minute tour ×
Programmers Stack Exchange is a question and answer site for professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.

I have an exam question that I am unsure of the answer. The question is:

In organisation X valid user names have the following structure. The user name can be either the employee’s name followed by a colon and then the department name or a string of one or more alphanumeric letters consisting of lower case, upper case letters and digits from 0 to 9. All user names end with a period ("."). The employee’s name is expressed as first name followed by an underscore and then the surname. The first name and the surname must begin with an uppercase letter and is followed by an arbitrary number of lower case letters. All department names are an arbitrary number of lower case letters. Express the language for valid user names as a regular expression.

The answer I have is,

[A-Z][a-z]+"_"[A-Z][a-z]+":"[a-z]+"." | [A-Za-z0-9]+"."

The next question is:

Construct a deterministic finite state automaton for recognising the user names as described in Question 1.

For which I have this:finite automata

I am under the impression this is wrong because [A-Z] is a subgroup of [A-Za-z0-9] and a DFA cannot have the same symbol going from one state. Does anybody have an idea how to solve this / is it correct?

Thanks !

share|improve this question
Wouldn't arbitrary number mean >=0 not >0? I.e. [A-Z][a-z]*.... –  Vadim Jan 3 '13 at 16:10
I thought so too but in that case a, "department name" could be nothing... –  Tom celic Jan 3 '13 at 16:38
+1 very interesting question –  Evan Plaice Jan 3 '13 at 17:08
add comment

3 Answers

As pointed out in the comments arbitrary can include 0, so you should have * instead of + in your regular expression at the corresponding places.

From your original regular expression you correctly derived a nondeterministic finite state automaton (NFA). To adapt it to the * mentioned above, you only need to eliminate the states 2 and 5 in order to have a corresponding NFA again.

The interesting part of the exam question though is to derive a deterministic finite state machine from your NFA. This is mechanically possible via an algorithm referred to as the powerset construction. This link should give you enough information (and you should have learned it in your lessons anways, given the exam question!) to learn how to get from NFA to DFA.

The resulting DFA is usually too large due to the powerset of states, which is the reason, why one normally performs a DFA minimization afterwards. Though to be honest, I would skip that given that the question only asks for a DFA, not a minimal DFA.

share|improve this answer
Correct me if I'm wrong but doesn't the Op need a valid NFA before he can determine a DFA? In this case, the NFA is invalid because it's impossible to determine whether the starting input is an employee_name or just a string in state 1 because [A-Z] is a subset of [A-Za-z0-9]. –  Evan Plaice Jan 3 '13 at 17:12
Of course it's a subset allowing you to use either transition. That's the whole point of non-determinism after all. –  Frank Jan 3 '13 at 22:24
OK, I think I see. If the first match fails then it backtracks to where it can evaluate the second (hence non-deterministic). I guess I have been mixing up NFA/DFA all along. That means my answer should outline how to convert the current NFA to a DFA. Guess I'm still a FSM noob. BTW +1 good answer. –  Evan Plaice Jan 4 '13 at 6:09
I suggest you read up on non-determinisim, backtracking, and probabilistic choice. Judging from your wording you have those all mixed up in your understanding (not sure about prob. choice, but from my experience most people who don't get non-det. can't distinguish it either) –  Frank Jan 4 '13 at 10:22
OK, read more. The way I described it is based on a PCRE (Perl Compatible Regular Expression) implementation. What you're referring to is something more like a Thompson NFA implementation where matches are checked in parallel and the best first match wins. I found this link (swtch.com/~rsc/regexp/regexp1.html) and mind == blown. –  Evan Plaice Jan 8 '13 at 3:37
add comment

I'm no expert but I think this assumption is correct.

I am under the impression this is wrong because [A-Z] is a subgroup of [A-Za-z0-9] and a DFA cannot have the same symbol going from one state.

It sounds like you have a pretty good grasp but I'll spell it out just to be thorough.

Following are the steps to convert the current NFA to a DFA. The first step is assessing where the non-determinism exist so the problem can be broken down further.

The initial split needs to happen based one of these two conditions:

The user name can be either the employee’s name followed by a colon and then the department name

Ie. an 'employee_name' name is encountered

a string of one or more alphanumeric letters consisting of lower case, upper case letters and digits from 0 to 9

Ie. a 'string' is encountered

It looks like you have identified the issue already. In that, there's no way to determine a split on state 1 because both paths share the common subset [A-Z][a-z]*. To create a valid NFA you first need to identify the constraints.

To start I'd look at separating out cases where the initial input is not an 'employee_name'.

Such cases include:

  • the first char is a lower-case alpha char
  • a numeric char is encountered

In addition, both an 'employee_name' and a 'string' can be made up of the identical subset [A-Z][a-z]*. To cover that condition there should be a two way split where either [] or [.] is encountered. Where [] is encountered you would transition what is currently state 4. Where [.] is encountered you'd transition to what is currently state 10.

Now that all of the constraints are identified, it's just a matter of re-mapping your state diagram to represent the flow. There should be one more state containing a * repeater for processing the rest of the 'string' value and 3 more transitions. It's just a matter of placing them correctly.

Following what @Frank said, you should also eliminate two of the states using * instead of + for repetition because the 1-or-more constraint is technically incorrect. For instance a valid user name could contain only one character of [A-Z] and no additional chars before the [_]. The same pattern applies to the 'surname'

share|improve this answer
add comment

from 1 to 9 remove the [A-Z] transition and add transitions form 2 and 3 to 9 with [A-Z0-9]

share|improve this answer
This really belongs as a comment to the question, unless you have more to add in which case you should expand upon your answer. –  Glenn Nelson Jan 3 '13 at 15:30
This does not provide an answer to the question. To critique or request clarification from an author, leave a comment below their post. –  Martijn Pieters Jan 3 '13 at 15:41
You really need to expound upon this answer or else it's going to be deleted for being too short to be constructive. Answers are expected to be comprehensive and explanatory. –  Jimmy Hoffa Jan 3 '13 at 17:05
add comment

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.