p-adic representations and period rings

-adic representations and period rings

Three rings for the Elven-kings under the sky,

Seven for the Dwarf-lords in their halls of stone,

Nine for the mortal Men doomed to die,

One ring to rule them all,

— P. Colmez

The goal of this post is to introduce the first layer of -adic Hodge theory: -adic Galois representations and Fontaine’s period rings. The guiding question is: what extra structure is carried by -adic Galois representations that arise geometrically?

Let be a finite extension, and let be its absolute Galois group. A -adic representation of is a finite-dimensional -vector space equipped with a continuous action of . Smooth proper varieties over produce such representations through their étale cohomology, .

The central paradigm of -adic Hodge theory is that these geometric representations carry Hodge-theoretic, de Rham, crystalline, and semistable structures. Like a prism splitting white light into its constituent colors, Fontaine’s period rings are the coefficient rings that reveal those structures. The path runs from -adic Galois representations, to invariants defined using period rings, to linear algebra carrying a filtration, a Frobenius, and a monodromy operator. This post focuses on the first two steps: the representations, and the period rings themselves.

A -adic Galois representation is a topological object: a continuous action of a profinite group on a -adic vector space. Algebraic de Rham cohomology is a linear-algebraic object: a filtered vector space over . , these live in different worlds, and there is no map between them until one enlarges the coefficients. The period rings are precisely the enlargement that lets the two worlds communicate. Each period ring is a -algebra carrying a -action together with some extra structure (a grading, a filtration, a Frobenius, a monodromy operator), and each is designed so that its invariants extract exactly one of those extra structures from a representation.

Later posts will study the linear algebra that emerges, in particular filtered -modules, weak admissibility, isocrystals, slopes, the construction of the Fargues–Fontaine curve, the classification of vector bundles on it, and the appearance of geometric local Langlands.

Local fields

We start by reviewing basic ideas on local fields.

Let be a finite extension. Its ring of integers is its maximal ideal is , and its residue field is . Since is finite, is a finite field, and we write . We choose a uniformizer , so that , and we normalize the valuation by . The ramification index and residue degree of are then defined by and . These two invariants recover the degree through the identity .

The ring is a discrete valuation ring: is a finite extension of the complete discretely valued field , and the valuation extends uniquely to with value group . Since is finite as a module over the complete DVR and is torsion-free, it is free, say of rank . We produce an explicit -basis and count.

Choose elements whose reductions form an -basis of ; this is possible because . We claim the elements form a -basis of . For spanning, take . Reducing modulo and lifting, we may write with representing residues and ; iterating on and collecting powers of that exceed into higher powers of (using , so ), the -adic completeness of lets the process converge and expresses as a -combination of the listed elements. For independence, a nontrivial relation of minimal -adic valuation among the coefficients would, after dividing by the appropriate power of , reduce to a nontrivial -relation among the in the graded pieces , contradicting the choice of the and the fact that have distinct valuations modulo . Hence the elements form a basis, and .

Let be the ring of Witt vectors of , and set . Then is the unique unramified extension of degree . After fixing an embedding , it is the maximal unramified subextension of , and is totally ramified. The field carries a Frobenius automorphism lifting the -power map on .

For a perfect field of characteristic , the Witt vector ring is the unique -adically complete discrete valuation ring with uniformizer and residue field ; this is the defining universal property of -typical Witt vectors. Since is finite of degree , the extension over is unramified (its uniformizer is still ) of residue degree , hence of degree , and it is the unique such extension because unramified extensions correspond bijectively to residue-field extensions.

The Frobenius is obtained by functoriality: the absolute Frobenius on is a ring homomorphism, and is a functor, so is a ring endomorphism of reducing to modulo . Since is finite, is an automorphism, so is an automorphism, and it inverts to . Inverting extends to .

Let be an algebraic closure of , and define , the completion of for the unique extension of the -adic absolute value. The field is complete and nonarchimedean by construction, and it is algebraically closed. The absolute Galois group is a profinite group, and its action on is isometric, hence extends continuously to .

The group is the inverse limit of the finite quotients as ranges over finite Galois extensions of inside , so it is profinite. Each preserves the absolute value on (the extension of to is unique, so it is Galois-invariant), hence is uniformly continuous and extends uniquely to the completion .

For algebraic closedness, we use Krasner’s lemma. Let be monic; we show it has a root in . Since is dense in , choose with small, and set . Then splits over , say with roots . The roots of and are close: if is any root of then is small, and since , some satisfies small. Making the approximation fine enough that is smaller than for every other root of , Krasner’s lemma gives ; since was already algebraic over and now lies within -distance tending to , the sequence of such converges in to , so . Hence every monic polynomial over has all roots in .

Inertia and Frobenius

Let be the maximal unramified extension of , and let be the inertia subgroup. Reduction to the residue field gives an exact sequence Since , the quotient is , topologically generated by the arithmetic Frobenius ; its inverse is the geometric Frobenius.

Let be the integral closure of in . It is a valuation ring (nondiscrete) with maximal ideal , and its residue field is : any element of is a root of a polynomial over , which lifts to a monic polynomial over whose roots lie in , so the residue field is algebraically closed over , hence equals . Each preserves and , so it induces an automorphism of fixing . The assignment is a continuous homomorphism .

The kernel consists of those acting trivially on . An automorphism acts trivially on the residue field if and only if it fixes every unramified subextension, that is, if and only if it lies in . Thus the kernel is .

For surjectivity, recall that finite unramified extensions of correspond bijectively to finite (necessarily separable, as is perfect) extensions of , with for the corresponding fields. Given any element of , its restrictions to finite lift to compatible elements of , and taking the inverse limit produces a preimage in . Finally, , with the topological generator , because the finite quotients are generated by compatibly in .

Thus, carries two fundamental types of information: inertia, measuring ramification, and Frobenius, measuring residue-field arithmetic. This dichotomy reappears throughout local arithmetic. In -adic cohomology for , unramified representations are largely controlled by Frobenius, and inertia acts through a finite quotient on a semisimplified representation. In -adic Hodge theory, the situation is more delicate, as inertia can act through infinite quotients in ways that encode Hodge filtrations and monodromy. Rendering precise that encoding is the purpose of the period rings.

-adic representations

A -adic representation of is a finite-dimensional -vector space equipped with a continuous action . Equivalently, it is a continuous homomorphism , where carries its -adic topology.

After choosing a basis of , a -adic representation of dimension is a continuous homomorphism . We denote the category of -adic representations of by . It is a tensor category: indeed, if , then so are , , , and .

For the direct sum, set . For the tensor product, set on simple tensors and extend -linearly; this is well defined because acts -linearly on each factor. For the dual, set , which is a left action because . For , this specializes to .

Continuity holds for each construction. Each of is a continuous functor on finite-dimensional -adic vector spaces (the relevant maps on matrix entries are polynomial in the entries of the constituent representations, and inversion is continuous on ), so composing and with these operations yields continuous homomorphisms.

The trivial representation is , with acting trivially.

Continuity is a substantive condition. Since is compact and is continuous, the image is a compact subgroup of . Compactness is what forces the existence of stable lattices, which in turn bind -adic representations to finite arithmetic objects.

Fix an isomorphism and let be the standard lattice, with stabilizer . The subgroup is open and compact in , so its preimage is open. Since is compact, has finite index; choose coset representatives for . Define a finite sum of -lattices, hence itself a -lattice in (finitely generated, and spanning over since it contains ).

We check -stability. Let . For each , the product lies in some coset, say , where is a permutation of (left multiplication permutes cosets). Writing with , and noting by definition of , we get . Therefore Thus is -stable.

Such a lattice allows one to approximate by the finite Galois modules . This is one reason -adic representations connect both to continuous -adic linear algebra and to finite arithmetic objects, and it is the mechanism behind reduction modulo of Galois representations.

The cyclotomic character

For every , let be the group of -th roots of unity. The group acts on , and after choosing a primitive root , every satisfies for a unique . The numbers are compatible as varies, and passing to the inverse limit produces the -adic cyclotomic character , characterized by for all -power roots of unity .

The group is cyclic of order , so an automorphism is determined by its action on a generator, and it must send a generator to a generator. Hence , and the Galois action gives a homomorphism . This homomorphism is continuous: its kernel is , which is open because is finite.

The transition maps , , are -equivariant, so the are compatible with the reduction maps . Taking the inverse limit gives a continuous homomorphism The defining identity holds level by level, hence for all -power roots of unity.

The representation is the one-dimensional -vector space on which acts by the cyclotomic character, . For , define , with the convention for .

These are the Tate twists. If is a -adic representation, define . Tate twists shift weights, and they appear naturally in cohomology because the cycle class of a subvariety of codimension lives in cohomology with a twist. The cleanest instance is projective space:

By the projective bundle formula in étale cohomology (applied to for an -dimensional vector space), the total cohomology ring is where is the first Chern class of , a canonical element of the -twisted . Thus is one-dimensional, generated by , and acts trivially on this generator (Chern classes are Galois-equivariant and is defined over ). Untwisting by gives . The odd cohomology vanishes because the ring is concentrated in even degrees.

For the base case , the identification can also be seen directly: the trace/degree map identifies with the dual of twisted by the cycle class of a point, and the Galois action on that class is by , matching the -adic Tate module of .

Thus, is the -adic cohomology class of a divisor, and it is the -adic analogue of the class in the complex comparison between Betti and de Rham cohomology. Recording where the twists sit is the content of Hodge–Tate theory below.

Representations from geometry

Let be a smooth proper variety. Then -adic étale cohomology produces finite-dimensional -vector spaces . Because acts on , it acts on the base change , and therefore on cohomology, making a -adic representation of .

For , the automorphism induces an automorphism of -schemes covering . Étale cohomology is a contravariant functor, so induces on , and the identity gives a group action (after passing to a left action via inverses). Passing to the limit over and inverting yields a -linear action on .

Continuity holds because , each finite group is a continuous discrete -module (its stabilizers are open, since the cohomology is already defined over some finite extension of ), and an inverse limit of continuous actions is continuous.

We refer to a -adic representation of as geometric, informally, if it occurs as a subquotient of for some smooth proper variety , some degree , and some Tate twist .

The precise meaning of “geometric” varies with context (the Fontaine–Mazur conjecture gives a purely Galois-theoretic characterization in the global setting). The relevant point here is that representations arising from geometry satisfy strong regularity properties, which the period rings are built to detect. More precisely, for smooth proper , the representation is always de Rham; if has good reduction it is crystalline; and if has semistable reduction it is semistable. Defining these three conditions is the work of the rest of the post.

Why can’t we just extend scalars?

Let be smooth and proper. There are two cohomology theories one would like to compare. The first is étale cohomology , a -vector space carrying a continuous -action. The second is algebraic de Rham cohomology, a -vector space carrying a decreasing Hodge filtration . The two carry the same numerical information (dimensions agree), and one wants a canonical comparison. Naively extending scalars, however, fails: has no canonical map in either direction.

We indicate somewhat imprecisely how the two sides carry incompatible structure. The left-hand side remembers a continuous -action; the right-hand side remembers differential forms and a filtration. Tensoring the left side with neither creates the Hodge filtration nor introduces differential forms, and it leaves a residual Galois action that the right side does not have. There is no functorial -linear identification that respects both structures.

The complex-analytic case sheds light on what is missing. For smooth proper, integration of algebraic forms over topological cycles gives period integrals , and assembling them into a period matrix induces a canonical isomorphism The field is large enough to contain all the periods, and in particular it contains , the period appearing in the comparison for of . The base field in the -adic setting has no room for such periods: it contains no analogue of . One therefore enlarges to a period ring: a -algebra with -action, large enough to contain the -adic periods, so that the comparison takes the form compatibly with the extra structure on both sides. Extracting -invariants will then recover each cohomology theory from the other.

The remainder of the post constructs the period rings , , , , and , in increasing order of the structure they see.

The completed algebraic closure

The simplest period ring is itself. It is a complete algebraically closed nonarchimedean field carrying a continuous -action. Its usefulness rests on a foundational theorem of Tate (with contributions of Ax and Sen), which computes its Galois invariants.

The inclusion is clear. The reverse inclusion is the heart of the matter and rests on the Ax–Sen estimate: there is a constant such that if is moved only slightly by , meaning for all , then lies within distance of . Applied to a genuinely invariant (where may be taken arbitrarily small by approximating by algebraic elements), the estimate forces to lie in the closure of inside , which is itself since is complete. We only indicate this; the quantitative approximation lemma is the technical core and is genuinely nontrivial.

For the twisted statement, the key input is Tate’s computation of the continuous cohomology of under , where . Tate’s normalized trace maps split (as a topological -module, up to the completed field ) into an invariant part and a part on which acts with no invariants for a nontrivial power of . A nonzero element of with would produce a line in on which , and hence , acts by while being pointwise fixed, which is impossible. Hence for . A complete account is Tate’s theorem on the Galois cohomology of .

Thus already detects Tate twists: an element of a -line is -fixed exactly when the twist vanishes. This single vanishing statement drives the entire theory of Hodge–Tate weights.

Hodge–Tate representations

Let be a -adic representation of . For each integer , define Because , each is a -vector space. There is a natural -linear comparison map

An element is definitionally a -invariant vector in . Tensoring with over and using the canonical isomorphism (the twist and its inverse cancel), the image of lands in . This defines a -linear map for each ; summing over gives . The map is -equivariant because is invariant and the identification is equivariant.

The representation is Hodge–Tate if is an isomorphism. The integers with are the Hodge–Tate weights of , with multiplicity .

With this convention, has Hodge–Tate weight .

We compute directly: By the Ax–Sen–Tate vanishing, this is unless . When we have , so the invariants are , one-dimensional. Hence the unique Hodge–Tate weight is , with multiplicity , and the comparison map is visibly an isomorphism of one-dimensional -spaces. In particular, is Hodge–Tate.

This is a deep comparison theorem, so we only indicate the shape of the argument. One first establishes it for building blocks where the periods are explicit: for and abelian varieties, the decomposition traces through Kummer theory and the Hodge decomposition of the tangent/cotangent spaces, and for projective space, it follows from the twist computation above.

In general, the theorem is obtained by working on the pro-étale site of with Fontaine’s period sheaves. That is, one constructs a filtered period sheaf whose cohomology computes after suitable twisting, proves a -adic Poincaré lemma identifying its de Rham complex with a resolution of the constant sheaf, and takes the associated graded. The graded of is , and the de Rham comparison (below) induces the Hodge–Tate comparison on associated gradeds. The precise argument is due to Fontaine, Messing, Faltings, and in the modern pro-étale form, to Scholze.

For instance, if , then , and the Hodge–Tate comparison reads Since is one-dimensional, this collapses to the tautology , consistent with the single Hodge–Tate weight .

The Hodge–Tate period ring

The Hodge–Tate decomposition can be packaged using the graded period ring which is a graded -algebra with -action (the multiplication uses ). For any -adic representation , set a graded -vector space whose degree- piece is . The representation is Hodge–Tate exactly when , the dimension on the left being the total dimension across all graded pieces.

Since and tensoring and invariants commute with finite direct sums, The comparison map is -linear between two -vector spaces of dimensions and . It is always injective (the general regularity argument given below for applies verbatim with in place of , using in degree ). An injective linear map of finite-dimensional spaces is an isomorphism precisely when the dimensions agree. Hence is Hodge–Tate if and only if .

The ring sees only the associated graded of the Hodge filtration. To recover the full filtration, one needs a period ring that remembers how the graded pieces are glued, namely . Constructing it requires descending to characteristic and back; we explain this shortly.

The tilt of

The more refined period rings are built from the tilt of , its characteristic- avatar. Define the inverse limit of along the -power (Frobenius) map. An element of is a sequence with in for all . The fraction field is a complete algebraically closed nonarchimedean field of characteristic .

Addition and multiplication are componentwise, and since each has characteristic , so does the inverse limit; thus is an -algebra. To define its valuation we use the untilting map constructed just below, setting ; this is multiplicative, and one checks it is a valuation with the same value group as (the -power roots make the group -divisible). Completeness of for this valuation follows from completeness of , since a Cauchy sequence in is a compatible system of Cauchy sequences in . Algebraic closedness is the tilting equivalence: finite extensions of correspond functorially to finite extensions of , and is algebraically closed, so has no nontrivial finite extensions. (Here, we treat the tilting equivalence as an external input, as it is the foundational theorem of perfectoid theory.)

The path back to characteristic is the multiplicative map . Given , choose arbitrary lifts of , and define

The key elementary fact is the lifting estimate: if in with , then . Indeed where , and each summand satisfies (since ), so the sum is . The product then lies in .

Now, the compatibility in means . Raising to the -th power and applying the estimate times upgrades this to Hence, is Cauchy in the -adically complete ring and converges. If is another choice of lifts, then , so the same estimate gives , and the two limits coincide. Thus, is well defined and independent of lifts.

The map is multiplicative but not additive.

Multiplicativity is immediate from and passage to the limit, since one may lift by the products . Additivity fails because addition in is the limiting operation which differs from in general: the binomial expansion of contains cross terms that do not vanish -adically before the limit is taken. Concretely, over one already sees for suitable non-integer tilts.

This construction is the first appearance of the characteristic- shadow of . It offers a glimpse of perfectoid tilting and, eventually, of the Fargues–Fontaine curve.

The ring

Define , the ring of -typical Witt vectors of . Since is perfect (Frobenius is bijective, being the shift on the tilting inverse limit), is -torsion-free and -adically complete, and every element has a unique Witt expansion where denotes the Teichmüller lift. There is a canonical surjective ring homomorphism , determined by and hence on Witt expansions by

It is instructive to explain why this is a morphism of rings. :::{.proof-like name=“Explanation.”} The formula is dictated by the universal property of Witt vectors. For a perfect -algebra , the ring is the unique -adically complete, -torsion-free ring with , and it is initial among such -adic thickenings. The pair presents as a -adic thickening of modulo (indeed ), so the universal property produces a unique ring homomorphism lifting the identity modulo . On Teichmüller representatives this map is forced to be , and additivity together with -adic continuity extends it to all Witt expansions. The Witt addition and multiplication laws are exactly compatible with this because they are engineered to make a functor to -adic thickenings. :::

The kernel is a principal ideal, generated by a non-zero-divisor. We fix a generator .

Choose a compatible system of -power roots of (possible since is -divisibly rich and is algebraically closed), so that . Then satisfies , so . In the cyclotomic variant one takes for the system of -power roots of unity introduced below; a short computation gives . That such a generates the whole kernel, and is a non-zero-divisor, is the statement that realizes as a perfectoid untilt: the kernel of for the untilt of a perfectoid ring is invertible, and over invertible ideals with the relevant reduction are principal. We take this principality as external input, since a self-contained proof requires the deformation theory of perfectoid rings.

The ring carries a Frobenius induced by the Witt-vector Frobenius, satisfying . The group acts on , hence on , hence on and on by functoriality, and the map is -equivariant (it is built from the equivariant ). Thus, already carries the three structures that drive the rest of the plot: a -action, a Frobenius , and the specialization .

The de Rham period ring

The ring is defined as the -adic completion of , The map extends to a surjection , and is a complete discrete valuation ring with residue field and maximal ideal ; any generator of is a uniformizer.

To produce the period of the Tate twist, choose a compatible system of -power roots of unity, , with primitive. Since , the Teichmüller element is a unit in mapping to , and . Define

Set . Since , we have , so and in the -adic topology of . The denominators are units in (we have inverted ), and their -adic valuations grow only logarithmically in , so as well. Because is -adically complete, the series converges. Thus , and in fact generates : modulo one has , and is a uniformizer, so is too.

The element transforms by the cyclotomic character.

Since compatibly, we have in , where the exponent acts through the -module structure of the -divisible group . Applying the Teichmüller lift and continuity, . Therefore where the identity holds for by continuity from the case .

Thus, is the -adic analogue of : it spans a -stable line on which the group acts by , exactly as spans the period of . Now define This is a complete discretely valued field with a decreasing, exhaustive, separated filtration , and its associated graded pieces are Tate twists of .

Multiplication by is an isomorphism . Since generates the maximal ideal of the DVR , the quotient is the residue field , with its natural -action. The isomorphism above is not -equivariant on the nose: it multiplies by , and , so it intertwines the action on with the action twisted by . Therefore, as a -module, .

Consequently . This gives a relationship between the de Rham and Hodge–Tate period rings.

We show that the ring has a continuous -action with .

The inclusion is clear. Conversely, let be nonzero, and let be its valuation, so . The image of in is a nonzero -invariant element, hence lies in . By Ax–Sen–Tate, for and for ; since , we must have and . Subtracting a lift in of (using , which exists because contains hence , and its residue map to restricts to the inclusion ) strictly raises the valuation of the invariant element . Iterating and using completeness of shows the successive corrections converge, expressing as an element of . Hence .

De Rham representations

Let be a -adic representation of . Define Because , this is a -vector space. It inherits a decreasing filtration from ,

The representation is de Rham if .

The following bound holds for every , and being de Rham means it is attained.

It suffices to show that the natural -linear map is injective; comparing -dimensions then gives . Equivalently, we show that any -linearly independent family remains -linearly independent in .

Suppose not, and take a nontrivial dependence relation with of minimal length among all such relations (reindexing so the involved indices are ). Since is a field we may normalize . Apply any . Because each is invariant, acts only on the coefficients: Subtracting the original relation and using kills the first term: This is a relation of length , so by minimality it is trivial, giving for all and all . Hence . But then is a nontrivial -linear relation among , contradicting their -linear independence. Therefore no such -relation exists, the map is injective, and the bound follows.

The same argument, replacing by , gives the analogous bounds for and below; this is the general phenomenon that , , and are -regular period rings.

We merely indicate the mechanism; the full theorem is one of the central results of the subject. On the pro-étale site of , Fontaine’s filtered period sheaf interpolates between the constant sheaf and the de Rham complex. A -adic Poincaré lemma states that the de Rham complex of (with its connection) resolves , so the pro-étale cohomology of is computed by . Comparing with the étale side, where recovers , yields the comparison isomorphism, filtered because is filtered. Properness gives finiteness and Poincaré duality, while smoothness gives the Poincaré lemma. Taking -invariants and using recovers . This is due to Faltings, with the pro-étale formulation due to Scholze.

Passing to associated gradeds in the de Rham comparison and using recovers the Hodge–Tate comparison. In particular, every de Rham representation is Hodge–Tate.

The crystalline period ring

The ring sees the filtration but forgets the Frobenius, because the completion that defines it destroys the Frobenius on (the ideal is not Frobenius-stable). To retain the Frobenius, one completes more gently, using divided powers.

Starting from , let be the -adic completion of the divided-power envelope of with respect to . Let us explain what this means.

A divided-power structure on an ideal is a family of maps for behaving formally like , meaning , , and the expected addition and scalar rules, even when is not invertible in .

The divided-power envelope is the universal -algebra in which acquires a divided-power structure. Adjoining divided powers is exactly what crystalline cohomology requires, since crystalline thickenings are pro-nilpotent thickenings equipped with divided powers on their defining ideals. Define Here, already lies in , since the divided powers make the terms converge -adically. The ring carries a continuous -action, a Frobenius , and a -equivariant embedding .

The -action and the Frobenius on both preserve as a set closed under the ambient operations needed to extend divided powers ( preserves because and agree modulo on Teichmüller elements up to the divided-power corrections), so both extend to the divided-power envelope and then to after inverting and . For the embedding, the universal property of the divided-power envelope maps into any -adically complete algebra where has divided powers; qualifies, since becomes topologically nilpotent and the denominators are invertible after inverting . Inverting on both sides gives ; injectivity holds because is a domain mapping nontrivially into the field .

The Frobenius scales by .

Using and continuity of ,

Finally, the crystalline invariants are the maximal unramified subfield.

The invariants are contained in , so . The refinement to comes from Frobenius: an element of carries a well-defined action of , and one shows that the -invariants of are precisely the -relevant unramified periods, whose fixed field is the maximal unramified subextension rather than all of . The ramified part of does not survive because it is not visible to the Frobenius-compatible crystalline structure; only , which is and on which acts, is retained. A rigorous proof analyzes the fundamental exact sequences relating , , and together with Ax–Sen–Tate, and is part of Fontaine’s foundational study of .

Crystalline representations

Let be a -adic representation of , and define This is a -vector space, and the Frobenius on induces a -semilinear Frobenius (semilinear because acts nontrivially on the coefficient field through ).

The representation is crystalline if .

As in the de Rham case, one always has : the regularity argument above applies verbatim with and in place of and , so any -linearly independent family in stays -linearly independent in .

The embedding makes crystalline representations de Rham, and it identifies their invariants after extending scalars.

If is crystalline, then the comparison map is an isomorphism at the crystalline level, (injective by regularity, surjective because of equality of dimensions). Tensoring along gives Taking -invariants and using together with (the -structure is already -fixed) yields which has -dimension . Hence is de Rham, and acquires its filtration through the embedding , with the Frobenius on as extra structure that alone does not record.

The special fiber has crystalline cohomology over , a finitely generated -module with a -semilinear Frobenius; inverting gives a -module over . The generic fiber has -adic étale cohomology with -action. Good reduction means the smooth proper model connects them with no monodromy. The comparison is realized by a period morphism built from the crystalline nature of : the divided-power structure on is exactly what permits crystalline thickenings of over to communicate with the characteristic-zero étale cohomology. Now, compatibility with Frobenius is automatic, because is present on both and , and the filtration is recovered after passing to . This is the Fontaine conjecture , proved by Fontaine–Messing, Faltings, and others.

Thus, good reduction is reflected representation-theoretically by the crystalline property, i.e., the -adic representation remembers the crystalline cohomology of the smooth special fiber, together with its Frobenius.

The semistable period ring

Good reduction is a stringent condition. Many varieties have only semistable reduction, where the special fiber is a normal-crossings degeneration, whereupon Frobenius alone underdetermines the representation (as one must also remember the monodromy of the degeneration). The corresponding period ring is this instance is . It contains and carries an additional monodromy operator , together with the Frobenius , satisfying the commutation relation

One realizes by adjoining a single logarithmic period. Fix a compatible system of -power roots of , and adjoin a formal variable (its image in depends on a choice of branch of , and different choices differ by elements of ). Specifically, one may write a polynomial ring in over , with the Frobenius and monodromy determined by The construction is independent of the auxiliary choices up to canonical isomorphism.

It suffices to check the relation on and on powers , since both sides are additive and . On the operator vanishes (it is and is -independent), so both and vanish there. On with , while The two agree, so on all of . Interpreting as a logarithm, the relation is the algebraic manifestation of the fact that Frobenius rescales logarithmic monodromy by : if , then differentiating after applying picks up an extra factor of .

The ring thus carries rich structure by way of a -action, a Frobenius , a monodromy operator , and a -equivariant embedding (depending on the choice of branch of ). Its invariants are again , since acts on through , which contributes nothing new to the invariants beyond .

Semistable representations

For a -adic representation , define This is a -vector space equipped with a -semilinear Frobenius and a -linear monodromy operator , inherited from and satisfying .

The representation is semistable if .

As before, the regularity bound gives for every . The chain of embeddings furnishes the following hierarchy of representation classes.

Suppose is crystalline, so . Tensoring along gives ; taking invariants shows has full dimension, so is semistable (indeed with in this case).

Suppose is semistable, so . Tensoring along and taking -invariants (using ) gives of full -dimension, so is de Rham.

Suppose is de Rham, so . Passing to the associated graded with respect to the -adic filtration and using produces the Hodge–Tate comparison isomorphism , so is Hodge–Tate. This gives the full chain of implications.

The hierarchy is strict: indeed, there are Hodge–Tate representations that are not de Rham, de Rham representations that are not semistable, and semistable representations that are not crystalline, the last illustrated by elliptic curves with multiplicative reduction below.

Under semistable reduction the special fiber is a reduced normal-crossings divisor, which is smooth in the sense of logarithmic geometry once equipped with its natural log structure. Log-crystalline (Hyodo–Kato) cohomology replaces crystalline cohomology, and the log structure produces the monodromy operator , measuring the combinatorics of how the components of meet. The semistable comparison identifies the generic-fiber étale cohomology with the log-crystalline cohomology of the special fiber after extending scalars to ; the extra logarithmic period is exactly what encodes the monodromy. This is the conjecture , proved by Kato, Tsuji, and others.

Thus, semistable reduction is reflected representation-theoretically by the appearance of a nonzero monodromy operator.

A table of period rings

The period rings introduced above may be summarized as follows. The corresponding Fontaine functors are the invariants for , and the hierarchy of representation classes is

Example: Tate twists

Let , with basis satisfying . Recall . Then is -invariant in .

We compute directly, using multiplicativity of the actions on the two tensor factors: Hence, is invariant. Since is one-dimensional and is a unit in , this single invariant spans, so has full dimension , and is de Rham.

The same invariant works over (as ), giving , so is crystalline. Since , so Frobenius acts on the crystalline line by . The Hodge–Tate weight of is , as computed earlier. In particular is crystalline with Hodge–Tate weight and crystalline Frobenius eigenvalue , consistent with the geometric identification .

Example: projective space

Let . Then for , and the odd cohomology vanishes, as established above. On the de Rham side, is one-dimensional over , generated by the -th power of the hyperplane class, and the Hodge filtration places this generator in degree (it lives in but not ).

The standard model is smooth and proper, so has good reduction, and crystalline comparison applies: a -line on which Frobenius acts by (matching the computation for above). Thus, the single Tate twist simultaneously records the Hodge–Tate weight , the de Rham filtration jump in degree , and the crystalline Frobenius slope . This coincidence of three integers is the dictionary of -adic Hodge theory in its simplest incarnation: for a Tate twist, the Hodge–Tate weight, the filtration jump, and the Frobenius slope all agree.

Example: elliptic curves

Let be an elliptic curve. Its -adic Tate module is , and is a two-dimensional -adic representation of .

For each , the -torsion is a free -module of rank (an elliptic curve over an algebraically closed field of characteristic has ). The transition maps are multiplication by , which are surjective with the expected kernels, so the inverse limit is a free -module of rank . The Galois action on the coordinates of torsion points is continuous (each is defined over a finite extension), giving a continuous -action on , and inverting produces the two-dimensional -representation .

Étale cohomology recovers this, up to duality:

For a smooth proper curve, is canonically dual to the rational Tate module of the Jacobian: . For an elliptic curve the Jacobian is itself (via the canonical principal polarization sending to ), so and .

The Hodge–Tate comparison in degree reads and both -spaces on the right are one-dimensional.

An elliptic curve has genus , so (the space of invariant differentials). Serre duality on the curve gives , whence as well.

Therefore, has Hodge–Tate weights and , each with multiplicity . The reduction type of refines this into the finer structure detected by and .

If has good reduction, extends to a smooth proper elliptic scheme over , so crystalline comparison applies and is crystalline, with identified with the crystalline cohomology of the special fiber and its Frobenius (whose characteristic polynomial encodes the point count of mod ).

If has multiplicative (semistable, not good) reduction, then is semistable but not crystalline, and its monodromy operator is nonzero.

Multiplicative reduction is modeled analytically by a Tate curve for a parameter with . The rigid-analytic uniformization realizes as an extension where the sub comes from the roots of unity inside and the quotient from the lattice . This extension is nonsplit, and its class in is exactly (Kummer theory). Because the degeneration is logarithmic, the representation is semistable, and the nonsplitting corresponds precisely to a nonzero monodromy operator on , which maps the weight- line onto the weight- line and records the loop created by the nodal special fiber. Since , the representation is not crystalline, as crystalline representations have .

As we hinted at above, elliptic curves thus give the first legitimately nontrivial instance of the hierarchy: good reduction produces crystalline representations, while multiplicative reduction produces semistable representations that are not crystalline, distinguished exactly by the vanishing or nonvanishing of .

The geometric dictionary

The comparison theorems assemble into a dictionary between the geometry of and the linear-algebraic output extracted from , which we summarize in the following table. The functors , , and identify the period-ring invariants with actual geometric cohomology: and, under good reduction, The upshot is that a topological Galois representation has been resolved into filtered linear algebra over and , equipped with a Frobenius and a monodromy operator.

This is the entry point to the next post, which reverses the direction of inquiry. Rather than starting from a representation and extracting its invariants, one starts from the linear algebra, the filtered -modules over , and asks which of them actually arise from representations. The answer is Fontaine’s theorem that the essential image consists of the weakly admissible modules, and pursuing it leads to Hodge and Newton polygons, weak admissibility, and eventually the Fargues–Fontaine curve.

No comment found.

Add a comment

You must log in to post a comment.

Follow