An elementary proof of Cohen-Gabber theorem in the equal characteristic $p>0$ case

This paper presents a new, elementary proof of the Cohen-Gabber theorem specifically for the equal characteristic $p>0$ case.

The Gist

Imagine you are trying to understand a massive, complex, and slightly messy city. This city is called $A$. It has a specific layout (dimension $d$), but it's built on a strange foundation where the rules of arithmetic are different from what we use in daily life (this is the "equal characteristic $p > 0$" world, where numbers wrap around like a clock with $p$ hours).

The mathematicians in this paper, Kurano and Shimoto, are trying to solve a puzzle: Can we find a simple, perfect, grid-like blueprint hidden inside this messy city?

The Big Goal: The Cohen-Gabber Theorem

In the world of math, there is a famous theorem (Cohen's Theorem) that says: "Yes, you can always find a simple grid inside this messy city." Think of this grid as a perfect, orderly neighborhood made of power series (infinite polynomials).

However, the original theorem had a flaw. It said the messy city was built on top of this grid, but it didn't guarantee that the connection between the two was smooth. Sometimes, the city was built on the grid in a "twisted" or "separable" way that made it hard to study the city's structure using the grid.

The Cohen-Gabber Theorem is the upgrade. It says: "Not only can we find the grid, but we can also find a specific version of the grid where the city sits on top of it smoothly and cleanly (mathematically called 'separable')." This makes the city much easier to analyze.

The Problem: The "Twisted" Connection

Imagine you are trying to map a complex building ($A$) onto a simple grid ($k[[y_1, \dots, y_d]]$).

• The Old Way: You might pick a grid that fits, but the building is attached to it in a weird way. If you try to walk from the grid to the building, you might hit a wall or get stuck in a loop. In math terms, the "field extension" is not separable.
• The New Way: The authors prove you can rearrange the grid (by changing the coordinates slightly) so that the building sits perfectly on top of it, allowing you to walk freely between them without getting stuck.

How They Did It: The "Magic Wand" of Change

The paper provides a new, simpler proof for this. Here is the analogy of their method:

1. The Messy Room: Imagine the city $A$ is a room filled with $d+1$ or more pieces of furniture (variables). It's too crowded to see the clean grid.
2. The "Distinguished" Furniture: The authors look at the "tallest" pieces of furniture (mathematically, "distinguished polynomials"). These are the pieces that define the shape of the room.
3. The Problem with Derivatives: In this strange math world (characteristic $p$), if you try to measure the slope of a wall (take a derivative), sometimes the slope becomes zero even if the wall is there. This is the "twist" that makes things messy.
4. The Solution (The "Shift"): The authors show that if you have a wall with a zero slope, you can shift the furniture.
   • Imagine you have a wall defined by $X^p + Y$. Its slope with respect to $X$ is zero (because in this math world, the derivative of $X^p$ is 0).
   • The authors say: "Let's rename our coordinates!" Instead of measuring from $X$, let's measure from $X + \text{something}$.
   • By mixing the variables slightly (like adding a tiny bit of $Y$ to $X$), they change the equation so that the slope is no longer zero.
5. The Result: Once the slope is non-zero, the wall is "separable." The connection is now smooth. They prove you can always do this shifting trick until the whole city is sitting on a smooth, clean grid.

Why Does This Matter?

Think of the city $A$ as a complex machine. To fix it or understand how it works, you need to take it apart.

• If the machine is built on a "twisted" grid, taking it apart is a nightmare; parts might break or fuse together in weird ways.
• If the machine is built on a "separable" grid (the Cohen-Gabber result), you can take it apart piece by piece, knowing exactly how each part relates to the next.

This new proof is "elementary" because it doesn't use heavy, obscure machinery. Instead, it uses clever, step-by-step rearrangements (like the furniture shifting) to show that the perfect grid is always there, waiting to be found.

Summary in a Nutshell

• The Problem: Complex mathematical structures are hard to study because they are built on "twisted" foundations.
• The Solution: We can always find a "smooth" foundation (a coefficient field) by slightly adjusting our perspective (changing coordinates).
• The Analogy: It's like realizing that a crooked painting isn't crooked because the wall is bad, but because you are looking at it from the wrong angle. Once you shift your angle (the proof), the painting hangs perfectly straight, and you can finally appreciate the art.

This paper gives us a new, simpler way to prove that we can always find that perfect angle.

Technical Summary

1. Problem Statement and Context

The Cohen Structure Theorem:
Classically, Cohen's structure theorem states that any complete local ring $(A, \mathfrak{m}, k)$ of equal characteristic $p > 0$ contains a coefficient field $\phi: k \to A$ (a section of the quotient map $\pi: A \to k$) and a system of parameters $x_1, \dots, x_d$ such that $A$ is a module-finite extension of the formal power series ring $\phi(k)[[x_1, \dots, x_d]]$.

The Gap (Cohen-Gabber Theorem):
While Cohen's theorem guarantees a module-finite extension, it does not guarantee that the extension is "nice" in terms of field theory. Specifically, for the extension to be useful in étale cohomology and Bertini-type theorems, the induced field extension on the fraction fields of the minimal primes must be separable.

The Cohen-Gabber Theorem (proved by Gabber) strengthens Cohen's result: It asserts that one can choose the coefficient field and the system of parameters such that for any minimal prime $P$ of $A$ with $\dim(A/P) = d$, the field extension
$$\text{Frac}(\phi(k)[[y_1, \dots, y_d]]) \to \text{Frac}(A/P)$$
is separable.

The Goal:
The authors aim to provide a new, elementary proof of this theorem specifically for the equal characteristic $p > 0$ case. Previous proofs (by Gabber) relied on more advanced machinery; Kurano and Shimomoto seek to derive this using only basic commutative algebra, the Weierstrass Preparation Theorem, and properties of derivations.

---

2. Methodology

The proof is structured inductively and relies heavily on manipulating the choice of the coefficient field and the system of parameters to ensure separability.

Key Tools Used:

1. Reduction to the Reduced Equi-dimensional Case:
   The authors first reduce the general case to the case where $A$ is a reduced, equi-dimensional complete local ring. If the theorem holds for $A/\bigcap P_i$ (where $P_i$ are minimal primes of maximal dimension), it can be lifted back to $A$ using Cohen's existence theorem for coefficient fields.
2. Weierstrass Preparation Theorem:
   Used to represent elements in power series rings as units times distinguished polynomials. This allows the authors to treat the ring $A$ locally as a quotient of a power series ring $R = k[[X_1, \dots, X_{d+1}]]$ by a single element (or product of elements) $f$.
3. Derivations and Separability:
   A key insight is that a finite extension of fields in characteristic $p$ is separable if and only if the derivative of the minimal polynomial is non-zero. The authors use partial derivatives ($\frac{\partial f}{\partial X_i}$) to detect inseparability.
4. Change of Variables (Coordinate Transformation):
   The core of the "elementary" proof involves constructing specific coordinate changes (replacing variables $X_j$ with $X_j - X_1^n$) and modifying the coefficient field (using $p$-bases) to ensure that the partial derivative of the defining polynomial with respect to a specific variable is non-zero.

Proof Strategy:

The proof proceeds by induction on the number of generators of the maximal ideal $\mathfrak{m}$ beyond the dimension $d$.

• Base Case ($h=0$): $A$ is already a power series ring.
• Base Case ($h=1$): The ring is a hypersurface $A \cong k[[X_1, \dots, X_{d+1}]]/(f)$. This is the hardest part of the proof (Proposition 3.1).
• Inductive Step ($h \ge 2$): The authors construct a diagram of subrings, applying the $h=1$ case to a subring and then using the induction hypothesis on a smaller subring to build the final coefficient field and parameters.

---

3. Key Contributions and Technical Results

A. The "Hardest Case" (Proposition 3.1)

The authors prove the theorem for the case where $\mathfrak{m}$ is generated by $d+1$ elements.

• Setup: $A \cong R/(f)$ where $R = k[[X_1, \dots, X_{d+1}]]$ and $f = f_1 \cdots f_r$ (irreducible factors).
• The Challenge: We need to ensure $\frac{\partial f_i}{\partial X_1} \neq 0$ (or a similar variable) to guarantee separability. If the derivative is zero, the extension is purely inseparable.
• The Solution (Claim 3.2): The authors show that by changing the coefficient field and replacing variables, one can force the partial derivative to be non-zero.
  • Step 1 (Variable Change): If $\frac{\partial f_s}{\partial X_j} \neq 0$ for some $j \neq 1$, they replace $X_j$ with $X_j - X_1^n$ for a sufficiently large $n$. This shifts the terms in the polynomial expansion such that the derivative with respect to $X_1$ becomes non-zero.
  • Step 2 (Coefficient Field Change): If $\frac{\partial f_s}{\partial X_j} = 0$ for all $j$, then $f_s$ has coefficients not in $k^p$. They use the existence of a $p$-basis to modify the coefficient field $\phi: k \to A$ via a map $\psi_\delta$ that introduces a linear term in $X_1$ into the coefficients. This ensures the derivative becomes non-zero.

B. Inductive Construction (Proof of Theorem 1.1)

For the general case where $\mathfrak{m}$ has $d+h$ generators ($h \ge 2$):

1. They isolate a subring $B = \phi(k)[[x_1, \dots, x_{d+1}]]$.
2. Apply the $h=1$ result (Proposition 3.1) to $B$ to find a new coefficient field $\phi'$ and parameters $y_1, \dots, y_d$ such that $B$ is separably finite over $C = \phi'(k)[[y_1, \dots, y_d]]$.
3. They construct an intermediate ring $D$ and apply the induction hypothesis to find a final coefficient field $\phi''$ and parameters $z_1, \dots, z_d$ for a subring $E$.
4. By analyzing the tower of field extensions, they prove that the composition of separable maps remains separable, satisfying the theorem for $A$.

C. Counter-Example Illustration (Example 3.3)

The paper concludes with an example showing why the choice of the coefficient field is critical.

• In the ring $A = F_p(t)[[X, Y]]/(tX^p + Y^p)$, if one chooses the coefficient field $F_p(t)$, the extension is not separable (the derivative is zero).
• However, by choosing a different coefficient field $F_p(s)$ where $s = t+X$, the ring becomes $F_p(s)[[X, Y]]/((s-X)X^p + Y^p)$, and the extension becomes separable. This illustrates the necessity of the "flexibility" in choosing the coefficient field provided by the theorem.

---

4. Significance

1. Elementary Nature: The proof avoids heavy machinery (such as alterations or deep scheme-theoretic results) often associated with Gabber's original work. It relies on standard commutative algebra tools (Weierstrass preparation, derivations, $p$-bases), making the result more accessible to a broader audience of algebraists.
2. Applications: The Cohen-Gabber theorem is a foundational tool for:
   • Gabber's Alteration Theorem: Used extensively in étale cohomology to resolve singularities.
   • Bertini-type Theorems: Essential for proving the existence of smooth hyperplane sections in positive characteristic.
   • Perfectoid Spaces: While not explicitly mentioned, techniques related to coefficient fields and separability in characteristic $p$ are relevant to modern developments in arithmetic geometry.
3. Constructive Insight: The proof provides a constructive method (via coordinate changes and coefficient field adjustments) to "fix" inseparable extensions, offering a clearer understanding of the geometric structure of complete local rings in characteristic $p$.

In summary, Kurano and Shimomoto provide a rigorous, self-contained, and elementary demonstration that every complete local ring of equal characteristic $p > 0$ admits a "good" presentation as a finite separable extension of a power series ring, resolving a subtle but crucial point in the structure theory of local rings.