Let $R$ be a perfect $\mathbb{F}_p$-algebra and write $W(R)$ for the Witt ring [i.e., ring of Witt vectors -- PLC] on $R$. I want to know how much we can deduce about $\text{Spec } W(R)$ from knowledge of $\text{Spec } R$. I am especially interested in the case that $R$ is a domain.

Suppose that $R$ is a field. Then it's not hard to see that $W(R)$ is a DVR, so in this case the answer is completely known. For a general $R$, surjections $R \rightarrow R/\mathfrak{p}$ lift to surjections $W(R) \rightarrow W(R/\mathfrak{p})$, and as the latter is a domain, we get a copy of the spectrum of $R$ inside the spectrum of $W(R)$.

Sometimes, though, there are extra primes in the spectrum of $W(R)$! Here's the case that motivated my question. (These rings come up in the study of $(\phi, \Gamma)$-modules and in the construction of Fontaine's rings of periods.) Let $R$ be the "perfection" of $\mathcal{O}_{\mathbb{C}_p}/p$. That is, $R$ is the set of sequences $(x_i)$, $i \geq 0$, where each $x_i \in \mathcal{O}_{\mathbb{C}_p}/p$ and $x_i^{p} = x_{i-1}$. Note that $R$ is a (non-discrete!) valuation ring; reducing mod the maximal ideal of $\mathcal{O}_{\mathbb{C}_p}$ and projecting to the first component gives a surjection to $\overline{\mathbb{F}_p}$. It's not hard to show $R$ is a domain.

There's three obvious prime ideals. First, as before, there's the prime ideal $(p)$, the kernel of the surjection $W(R) \rightarrow R$. Further, the universal property of Witt vectors gives a map $W(R) \rightarrow W(\overline{\mathbb{F}_p}) = \widehat{\mathcal{O}_{\mathbb{Q}_p^{ur}}}$ whose kernel is a prime. Finally there's the maximal ideal, which is the kernel of $W(R) \rightarrow R \rightarrow \overline{\mathbb{F}_p}$. We've used only abstract reasoning about Witt vectors to find these.

But there's a fourth prime ideal! It turns out that there's a surjection $\theta: W(R) \rightarrow \mathcal{O}_{\mathbb{C}_p}$, which is critical in building Fontaine's rings. The existence of $\theta$ can't be deduced from the universal property of Witt vectors, since $\mathcal{O}_{\mathbb{C}_p}$ is not a strict $p$-ring. In the literature, all proofs that $\theta$ is a homomorphism "look under the hood" and actually think about the addition and multiplication of Witt vectors. Can this be abstracted? That is, is there a way to know which "extra" primes we'll get in $W(R)$, just from knowing $R$?

  • $\begingroup$ No thoughts on how to answer your question, but as I think I know what brought it up let me ask: is there any ergodic reasons behind this question? $\endgroup$
    – Ben Weiss
    Mar 7, 2010 at 6:13
  • $\begingroup$ No, I don't think so. $\endgroup$ Mar 7, 2010 at 16:11
  • 11
    $\begingroup$ This is not an answer to your question but just a comment on your comments about $\theta$. From my point of view, a better definition of $W(R)$, where $R$ is the perfection of $O_{\mathbf{C}_p}$ is the inverse limit of $W(O_{\mathbf{C}_p})$ with respect to the Frobenius maps $F$. (It's then an exercise to prove that this agrees with $W(R)$.) The map to $O_{\mathbf{C}_p}$ is then just the projection onto the zeroth component of the zeroth Witt vector in the system.It's surjective because for any $x\in O_{\mathbf{C}_p}$, the sequence of Teichm\"uller elements $[x],[x^{1/p}],\dots$ is in it. $\endgroup$
    – JBorger
    Apr 24, 2010 at 9:18
  • 4
    $\begingroup$ Wait, wait -- you mean the "ring of Witt vectors", not the "Witt ring in the sense of quadratic forms", as became clear when I read farther. Mea culpa. (Witt has too many rings!) $\endgroup$ Aug 25, 2010 at 9:47
  • 19
    $\begingroup$ Thanks for the edit! There should really be a more succinct name for "the Witt vector ring over R," given that brevity is the soul of Witt. $\endgroup$ Aug 25, 2010 at 14:44

1 Answer 1


This might not be the answer you expect, but there is a result saying that when $R$ is a non-discrete valuation ring, like the $\mathcal{O}_\mathbb{C_p}^\flat$ you mentioned, then there are infinitely many prime ideals in $W(R)$. The following two papers show that the Krull dimension of $W(R)$ is infinite, moreover, at least the cardinality of the continuum.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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