Should I cite a paper that I haven't read if I use a theorem in it?
It depends how well-known and wide-spread the result is. If by now it is a very well-known result and the publication already happened quite some time ago, it may not be necessary to quote the original paper. In doubt I'd quote the article and the books, or at least one of the books. A possible formulation could be:
The result below is due to Paul [reference to article]; we recall it in the form given in a monograph by Miller [reference to book].
I do not think it is necessary to have read the original paper to be entitled to write this. If it is not too much work to track it down, it could still make sense to at least take a peek at the article you quote.
When I cite a theorem, I typically have three goals.
- To point the reader to a precise statement and proof of the result.
To accomplish this goal, I cite the clearest source I know of, or the one best-suited to the way I'm using the theorem. In my experience, the clearest source is often not the original one. As people use a result, they often notice places where the statement or proof can be clarified. As a result, the presentation of a theorem often improves with age, even if its mathematical content remains the same.
- To point the reader to a trusted statement and proof of the result.
I must admit that I've cited many theorems whose proofs I don't fully understand—especially considering that, to really understand a proof, you also need to understand the results it depends on, and the results those results depend on... If you have read and understood a proof of the theorem you want, though, I might encourage you to cite that proof, on the principle that the best proof is a trusted proof. As a bonus, since clear proofs are by definition easier to read than confusing ones (in most circumstances, at least), this tends to support goal 1.
- To give the prover credit for the result, if necessary.
In some cases, the attribution of a theorem is already carried in its name, so there's no need to use a citation to give credit. In other cases, however, the history of a theorem can be hard to track down, so it's nice to cite a proof by the people the theorem is attributed to. Unfortunately, this often conflicts with goal 1, since the first way something was done is rarely the best way it can be done. A nice solution, if available, is to cite a later, cleaned-up proof written by the result's original authors. Another solution is to cite the clearest reference up front, but add an attribution citation later in the sentence, or in a footnote.
You should read the original paper and then cite it.
Citations exist for many reasons.
You should cite articles to show that you've done your due-diligence in learning the background information. Like the infamous Van Halen brown M&Ms, this is a quick way for your readers and reviewers to check that you actually know what you're talking about.
You cite other articles to allow readers to find the relevant background material. Research papers create a web of information and providing signs pointing to the important previous work is important for those not immersed in your paper's field.
Finally, you cite other work to give recognition and "props" to those who have done important work in the past. This is especially important to any work that you're actively building off of.
All three reasons I mention above are relevant to your question. This is your Master's thesis. You need to show that you've down the background reading, give an overview for future readers, and cover the important previous work. Sure you can cover secondary sources, but primary sources are better. If you're already familiar with the theorem and its proof, skimming the original paper shouldn't take much time and citing it is the more correct thing to do. Doubly so for a thesis.