aa r X i v : . [ m a t h . C O ] S e p Witt vectors, semirings, and total positivity
James Borger ∗ Abstract.
We extend the big and p -typical Witt vector functors from commutative ringsto commutative semirings. In the case of the big Witt vectors, this is a repackaging of somestandard facts about monomial and Schur positivity in the combinatorics of symmetricfunctions. In the p -typical case, it uses positivity with respect to an apparently newbasis of the p -typical symmetric functions. We also give explicit descriptions of the bigWitt vectors of the natural numbers and of the nonnegative reals, the second of whichis a restatement of Edrei’s theorem on totally positive power series. Finally we givesome negative results on the relationship between truncated Witt vectors and k -Schurpositivity, and we give ten open questions. Primary 13F35, 13K05; Secondary 16Y60,05E05, 14P10.
Keywords.
Witt vector, semiring, symmetric function, total positivity, Schur positivity.
Contents N , the general theory 2811.1 The category of N -modules . . . . . . . . . . . . . . . . . . . . . . 2811.2 Submodules and monomorphisms . . . . . . . . . . . . . . . . . . . 2811.3 Products, coproducts . . . . . . . . . . . . . . . . . . . . . . . . . . 2811.4 Internal equivalence relations, quotients, and epimorphisms . . . . 2821.5 Generators and relations . . . . . . . . . . . . . . . . . . . . . . . . 2821.6 Hom and ⊗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2821.7 N -algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2831.8 Commutativity assumption . . . . . . . . . . . . . . . . . . . . . . 2831.9 A -modules and A -algebras . . . . . . . . . . . . . . . . . . . . . . . 2831.10 Hom A and ⊗ A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2841.11 Limits and colimits of A -modules and A -algebras . . . . . . . . . . 2841.12 Warning: kernels and cokernels . . . . . . . . . . . . . . . . . . . . 2841.13 Base change, induced and co-induced modules . . . . . . . . . . . . 2841.14 Limits and colimits of A -algebras . . . . . . . . . . . . . . . . . . . 2841.15 Base change for algebras . . . . . . . . . . . . . . . . . . . . . . . . 2851.16 Flat modules and algebras . . . . . . . . . . . . . . . . . . . . . . . 2851.17 Examples of flat modules . . . . . . . . . . . . . . . . . . . . . . . 285 ∗ Supported the Australian Research Council under a Discovery Project (DP120103541) anda Future Fellowship (FT110100728). J. Borger N N . . . . . . . . . . . . . . . . . . . . . . . 2872.7 Extending fibered categories to nonaffine schemes . . . . . . . . . . 2882.8 Additively idempotent elements . . . . . . . . . . . . . . . . . . . . 2882.9 Cancellative modules . . . . . . . . . . . . . . . . . . . . . . . . . . 2882.10 Strong and subtractive morphisms . . . . . . . . . . . . . . . . . . 2892.11 Additively invertible elements . . . . . . . . . . . . . . . . . . . . . 2893 Plethystic algebra for N -algebras 2903.1 Models of co- C objects in Alg K . . . . . . . . . . . . . . . . . . . . 2903.2 Co- L -algebra objects in Alg K . . . . . . . . . . . . . . . . . . . . . 2903.3 Plethystic algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2913.4 Example: composition algebras and endomorphisms . . . . . . . . 2923.5 Models of co- C objects in Alg K . . . . . . . . . . . . . . . . . . . . 2923.6 Flat models of co- C -objects in Alg K . . . . . . . . . . . . . . . . . 2923.7 Models of composition algebras . . . . . . . . . . . . . . . . . . . . 2934 The composition structure on symmetric functions over N K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2944.3 Symmetric functions . . . . . . . . . . . . . . . . . . . . . . . . . . 2944.4 Remark: Λ N is not free as an N -algebra . . . . . . . . . . . . . . . 2944.5 Elementary and Witt symmetric functions . . . . . . . . . . . . . . 2954.8 Explicit description of a Λ N -action . . . . . . . . . . . . . . . . . . 2974.9 Example: the toric Λ N -structure on monoid algebras . . . . . . . . 2984.10 Example: the Chebyshev line . . . . . . . . . . . . . . . . . . . . . 2984.11 Flatness for Λ N -semirings . . . . . . . . . . . . . . . . . . . . . . . 2984.12 Example: convergent exponential monoid algebras . . . . . . . . . 2984.13 Example: convergent monoid algebras . . . . . . . . . . . . . . . . 2994.14 Remark: non-models for Λ Z over N . . . . . . . . . . . . . . . . . . 2995 The Schur model for Λ Z over N Sch . . . . . . . . . . . . . . . . . . . . . . . 3005.3 Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3015.4 Remark: Λ
Sch is not free as an N -algebra . . . . . . . . . . . . . . 3015.5 Remark: explicit description of a Λ Sch -action . . . . . . . . . . . . 3025.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3025.7 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3025.8 Frobenius lifts and p -derivations . . . . . . . . . . . . . . . . . . . 3025.9 Remark: the necessity of nonlinear operators . . . . . . . . . . . . 3035.10 Remark: composition algebras over number fields . . . . . . . . . . 3035.11 Remark: representation theory and K-theory . . . . . . . . . . . . 304 itt vectors, semirings, and total positivity N -algebras 3056.1 W and W Sch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3056.2 The ghost map and similar ones . . . . . . . . . . . . . . . . . . . . 3066.4 Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3076.5 Coordinates for Witt vectors of rings . . . . . . . . . . . . . . . . . 3076.6 Witt vectors for semirings contained in rings . . . . . . . . . . . . 3086.7 Example: some explicit effectivity conditions . . . . . . . . . . . . 3086.8 Coordinates for Witt vectors of semirings . . . . . . . . . . . . . . 3096.9 Topology and pro-structure . . . . . . . . . . . . . . . . . . . . . . 3096.10 Teichm¨uller and anti-Teichm¨uller elements . . . . . . . . . . . . . . 3106.11 The involution and the forgotten symmetric functions . . . . . . . 3106.12 Example: the map N → W Sch ( A ) is injective unless A = 0 . . . . . 3117 Total positivity 3117.2 Total positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3127.6 Remark: W ( R + ) and W Sch ( R + ) as convergent monoid algebras . . 3137.7 Remark: W ( R + ) and entire functions . . . . . . . . . . . . . . . . 3137.9 Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3147.10 Counterexample: W does not preserve surjectivity . . . . . . . . . 3147.11 Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3148 A model for the p -typical symmetric functions over N p -typical Witt vectors and general truncation sets . . . . . . . . . . 3148.2 Positive p -typical symmetric functions . . . . . . . . . . . . . . . . 3158.5 p -typical Witt vectors and Λ-structures for semirings . . . . . . . . 3178.6 Remark: a partition-like interpretation of the bases . . . . . . . . . 3188.7 Relation to the multiple-prime theory . . . . . . . . . . . . . . . . 3188.8 Some explicit descriptions of W ( p ) ,k ( A ) . . . . . . . . . . . . . . . . 3198.9 W ( p ) ,k ( A ) when A is contained in a ring . . . . . . . . . . . . . . . 3198.10 Counterexample: The canonical map W ( p ) ,k +1 ( A ) → W ( p ) ,k ( A ) isnot generally surjective . . . . . . . . . . . . . . . . . . . . . . . . . 3208.11 Semirings and the infinite prime . . . . . . . . . . . . . . . . . . . 3209 On the possibility of other models 32110 k -Schur functions and truncated Witt vectors 32110.1 k -Schur functions and Λ Sch k . . . . . . . . . . . . . . . . . . . . . . 32210.3 Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32310.4 Truncated Schur–Witt vectors for semirings . . . . . . . . . . . . . 32310.5 Counterexample: Λ Sch k is not a co- N -algebra object . . . . . . . . . 32311 Remarks on absolute algebraic geometry 32476 J. Borger
Introduction
Witt vector functors are certain functors from the category of rings (always com-mutative) to itself. They come in different flavors, but each of the ones we willconsider sends a ring A to a product A × A × · · · with certain addition and multi-plication laws of an exotic arithmetic nature. For example, for each prime p , thereis the p -typical Witt vector functor W ( p ) ,n of length n ∈ N ∪ {∞} . As sets, wehave W ( p ) ,n ( A ) = A n +1 . When n = 1, the ring operations are defined as follows:( a , a ) + ( b , b ) = ( a + b , a + b − p − X i =1 p (cid:18) pi (cid:19) a i b p − i )( a , a )( b , b ) = ( a b , a p b + a b p + pa b )0 = (0 , , . For n ≥
2, the formulas in terms of coordinates are too complicated to be worthwriting down. Instead we will give some simple examples: W ( p ) ,n ( Z /p Z ) ∼ = Z /p n +1 Z ,W ( p ) ,n ( Z ) ∼ = (cid:8) h x , . . . , x n i ∈ Z n +1 | x i ≡ x i +1 mod p i +1 (cid:9) . In the second example, the ring operations are performed componentwise; in par-ticular, the coordinates are not the same as the coordinates above. For anotherexample, if A is a Z [1 /p ]-algebra, then W ( p ) ,n ( A ) is isomorphic after a change ofcoordinates to the usual product ring A n +1 . This phenomenon holds for otherkinds of Witt vectors—when all relevant primes are invertible in A , then the Wittvector ring of A splits up as a product ring.The p -typical Witt vector functors were discovered by Witt in 1937 [46] andhave since become a central construction in p -adic number theory, especially in p -adic Hodge theory, such as in the construction of Fontaine’s period rings [16]and, via the de Rham–Witt complex of Bloch–Deligne–Illusie, in crystalline coho-mology [20]. Also notable, and related, is their role in the algebraic K-theory of p -adic rings, following Hesselholt–Madsen [19].There is also the big Witt vector functor. For most of this chapter, we willthink about it from the point of view of symmetric functions, but for the moment,what is important is that it is formed by composing all the p -typical functors inan inverse limit: W ( A ) = lim n W ( p ) , ∞ ( W ( p ) , ∞ ( · · · ( W ( p n ) , ∞ ( A )) · · · )) , (0.0.1)where p , . . . , p n are the first n primes, and the transition maps are given byprojections W ( p ) , ∞ ( A ) → A onto the first component. (A non-obvious fact is thatthe p -typical functors commute with each other, up to canonical isomorphism;so in fact the ordering of the primes here is unimportant.) This has an adelicflavor, and indeed it is possible to unify the crystalline cohomologies for all primes itt vectors, semirings, and total positivity p -adic integrality at the infinite prime. On the otherhand, we will have little to say about other aspects of the infinite prime, such asarchimedean norms or a Frobenius operator at p = ∞ .To explain this in more detail, we need to recall some basics of the theory ofsymmetric functions. The big Witt vector functor is represented, as a set-valuedfunctor, by the free polynomial ring Λ Z = Z [ h , h , . . . ]. So as sets, we have W ( A ) = Hom Alg Z (Λ Z , A ) = A × A × · · · . If we think of Λ Z as the usual ring of symmetric functions in infinitely many formalvariables x , x , . . . by writing h n = X i ≤···≤ i n x i · · · x i n , then the ring operations on W ( A ) are determined by two well-known coproductsin the theory of symmetric functions. The addition law is determined by thecoproduct ∆ + on Λ Z for which the power-sum symmetric functions ψ n = P i x ni are primitive, ∆ + ( ψ n ) = ψ n ⊗ ⊗ ψ n , and the multiplication law is determined by the coproduct for which the powersums are group-like, ∆ × ( ψ n ) = ( ψ n ⊗ ⊗ ψ n ) = ψ n ⊗ ψ n . This is also a fruitful point of view for the p -typical functors: W ( p ) , ∞ is rep-resentable by the free subring Λ Z , ( p ) = Z [ θ , θ p , θ p , . . . ] of p -typical symmetricfunctions, where the θ n ∈ Λ Z are the Witt symmetric functions, which are definedrecursively by the relations ψ n = X d | n dθ n/dd . The ring operations on the p -typical Witt functors are equivalent to coproducts onΛ Z , ( p ) , and these are compatible with the two coproducts on Λ Z . In fact, Λ Z canbe reconstructed from all the Λ Z , ( p ) as a kind of Euler product:Λ Z = Λ Z , (2) ⊙ Λ Z , (3) ⊙ Λ Z , (5) ⊙ · · · , (0.0.2)where ⊙ is the operation that on representing objects corresponds to composi-tion on functors. This is in fact just another expression of formula (0.0.1), oralternatively of Wilkerson’s theorem [45].78 J. Borger
Now, the relation of all this to the infinite prime is that there is a well-knownpositivity structure on Λ Z . This is the subset Λ N consisting of symmetric functionsthat have nonnegative coefficients when viewed as series in the formal variables x , x , . . . . It is closed under addition and multiplication and contains 0 and 1;so it is a semiring, or more plainly, an N -algebra. It is also well known that thecoproducts ∆ + and ∆ × above are induced by coproducts on Λ N , and so one mighthope to use them to extend the big Witt construction to all N -algebras, and henceto incorporate some positivity information in the usual theory of Witt vectors.This is indeed possible and the primary purpose of this chapter is to write it alldown in some detail.In fact, there is another such model over N , which is also well known. It isthe sub- N -algebra Λ Sch consisting of symmetric functions which have nonnegativecoordinates with respect to the basis of Schur symmetric functions.
Theorem A.
The functors
Hom
Alg N (Λ N , − ) and Hom
Alg N (Λ Sch , − ) extend the bigWitt vector functor W from Z -algebras to N -algebras. Each has a unique comonadstructure compatible with that on W . In terms of actual mathematical content, this is just a repackaging of somestandard positivity facts in the theory of symmetric functions. Thus a large part ofthis chapter is expository. Its goal is to convince the reader that there is interestingmathematics where Witt vectors, combinatorial positivity, and semiring theorymeet. To this end, I have included a number of open questions, which I hope willserve to focus readers’ minds and stimulate their interest. Most of the questionsare precise, of the yes/no variety, and some are no doubt well within reach.To give an example of something from this territory, I will report one newobservation, which is that there is also a positive model for the p -typical Wittvector functors. Theorem B.
There is a representable comonad on the category of N -algebraswhich agrees with the p -typical Witt vector functor W ( p ) , ∞ on Z -algebras. As with theorem A, the representing object is given by a positivity conditionwith respect to a Z -basis, in this case of Λ Z , ( p ) . Write d p = − θ p = ( ψ p − ψ p ) /p .Consider the (finite) monomials of the form Y i,j ≥ ( ψ ◦ ip ◦ d ◦ jp ) m ij , where ◦ denotes the plethysm operation on Λ Z , and where m ij < p . Then thisfamily of monomials is a Z -basis for Λ Z , ( p ) . Its N -linear span Λ N , ( p ) is a sub- N -algebra of Λ Z , and the functor on N -algebras it represents admits a unique comonadstructure compatible with that of W ( p ) , ∞ . To my knowledge, this basis of Λ Z , ( p ) has not been considered before.In one way, the theory around theorem B is more satisfactory than that aroundtheorem A. This is that it also works for the p -typical Witt vectors of finite length.I initially hoped that bases of k -Schur functions of Lapointe–Lascoux–Morse (seethe book [25]) would allow us to define N -algebras of big Witt vectors of finite itt vectors, semirings, and total positivity N —isa natural and well-behaved formalism, both in general and in its applications toWitt vectors and positivity. It has gotten almost no attention from people workingwith scheme theory over Z , but it deserves to be developed seriously—and inde-pendently of any applications, which are inevitable in my view. Let me concludewith some words on this.Arithmetic algebraic geometry can be regarded as the study of systems of poly-nomials equations over Z . Such a system is equivalent to a presentation of a Z -algebra; so one could say that arithmetic algebraic geometry is the study of thecategory of Z -algebras. Of course, arithmetic algebraic geometers study manyother objects, such as nonaffine schemes over Z , line bundles over them, and so on,but let us consider these as objects of derived interest, as tools for understandingpolynomial equations. In fact, many such concepts are formally inevitable once weallow ourselves the category of Z -algebras and some general category theory.Let me recall how this works for algebraic spaces. The category of affine schemesis defined to be the opposite of the category of rings. It has a Grothendieck topologyon it, the fppf topology, where covers are given by fppf algebras, those that arefaithfully flat and finitely presentable. The category of all algebraic spaces over Z (a slight enlargement of the category of schemes) is what one gets by taking theclosure of the category of affine schemes under adjoining arbitrary coproducts andquotients by fppf equivalence relations.This is a completely formal process. (For instance, see To¨en–Vaqui´e [43, 44].)Given a category C that looks enough like the opposite of the category of rings anda well-behaved class of equivalence relations, we can produce an algebraic geometryfrom C by gluing together objects of C using the given equivalence relations. Inparticular, we can do this with the category of N -algebras and produce a categorythat could be called the category of schemes over N . This brings positivity intoalgebraic geometry at a very basic level. In arithmetic algebraic geometry today,and specifically in global class field theory, positivity is treated in an ad hoc manner,much it seems as integrality was before the arrival of scheme theory in the 1950s.On the other hand, as it appears to me, most people working on semiring theoryfollow a tradition close to general algebra or even computer science. Scheme theoryhas had little influence. As someone raised in that tradition, I find this unaccept-able. The category of rings is the same as the category of semirings equipped witha map from Z . In other words, one might say that arithmetic algebraic geometryis nothing more than semiring theory over Z . One would therefore expect an ac-tive interest in finding models over N , or the nonnegative reals, of as many of theobjects of usual algebraic geometry over Z as possible, just as one always tries tofind models for objects of classical algebraic geometry, such as moduli spaces, over Z . Yet such an effort seems to be nearly nonexistent. Perhaps one reason for thisis that most existing expositions of scheme theory begin by considering spectra ofprime ideals, and it is less clear how to mimic this approach over N . Or perhaps80 J. Borger people are more interested in designing foundations for specific applications, suchas tropical algebraic geometry, rather than developing general tools. Whatever thecase, it is important to get beyond this.So in the first two sections, I give a category-theoretic account of the verybasics of semiring theory and algebraic geometry over N . It is largely expository. Ihope it will demonstrate to people who are skeptical that the basic constructions ofscheme theory extend to N , and demonstrate to semiring theorists a point of viewon their subject that emphasizes macroscopic ideas, such as flatness, base change,and descent, more than what is common. So at least the general formalism over N can be brought up closer to that over Z .If arithmetic algebraic geometry provides the motivation here and semiringtheory provides the formalism, then algebraic combinatorics provides us with thepositivity results. These are needed to define Witt vectors of semirings that donot contain −
1; they could be viewed as the analogues at the infinite prime ofthe slightly subtle p -adic congruences needed to define the p -typical Witt vectorsfor rings that do not contain 1 /p . But I also hope that combinatorialists will findsomething fresh in our emphasis on Witt vectors rather than symmetric functions.While the two are equivalent, they often suggest different questions. For instance,the coproduct ∆ × on Λ Z has gotten much less attention than ∆ + . But these arejust the co-operations that induce the multiplication and addition operations onWitt vectors. Although it is not without interest to view Witt vectors only asabelian groups, the real richness of the theory and their role in arithmetic alge-braic geometry comes when we remember their full ring structure (or even better,their Λ-ring structure). So to a specialist in Witt vectors, ignoring ∆ × might feellike missing the whole point. Also, aspects of symmetric functions related to thefinite primes seem under-studied in the algebraic combinatorics community. Forinstance, the ring Λ Z , ( p ) = Z [ θ , θ p , . . . ] of p -typical symmetric functions is, it ap-pears to me, nearly unknown there. (The symmetric function θ p does appear inMacdonald [33] as − ϕ p on p. 120.)I would like to thank Thomas Lam and Luc Lapointe for some helpful cor-respondence, especially on k -Schur functions. I would also like to thank LanceGurney, Lars Hesselholt, and especially Darij Grinberg for making some observa-tions that I have included. All automated computation I did for this paper wasdone in the software system Sage [40], especially using the algebraic combinatoricsfeatures developed by the Sage-Combinat community [37].Finally, I should mention that Connes and Consani [12] have developed a theoryof Witt vectors for certain algebras over the Boolean algebra { , } . It would beinteresting to know if there is any relation with our theory. Conventions
The word positive will mean >
0, and negative will mean <
0. For any subset A of the field R of real numbers, we will write A + := { x ∈ A | x ≥ } . itt vectors, semirings, and total positivity N of natural numbers is { , , , . . . } . The ring Z p of p -adic integers islim n Z /p n Z , and the field Q p of p -adic numbers is Z p [1 /p ].For any category C , we will write C ( X, Y ) = Hom C ( X, Y ) for short.
1. Commutative algebra over N , the general theory The primary purpose of this section and the next one is to collect the definitionsand formal results of commutative algebra and scheme theory over N that we willneed. The reader is encouraged to skip them at first and refer back only whennecessary.A general reference to the commutative algebra is Golan’s book [17]. Whileeverything here is essentially the same, there are some small differences. For in-stance, I have preferred to drop the prefix semi wherever possible and do not wantto assume 0 = 1. I have also followed the categorical method and used its termi-nology, because it gives the development a feeling of inevitability that I think isabsent from the more element-centric approaches. N -modules. The category
Mod N of N -modules is by def-inition the category of commutative monoids, which we typically write additively.Thus an N -module is a set M with a commutative binary operation + = + M andan identity element 0 = 0 M , and a homomorphism M → P is function f : M → P such that f (0 M ) = 0 P and f ( x + M y ) = f ( x ) + P f ( y ) for all x, y ∈ M . As usual,the identity element is unique when it exists; so we will often leave to the readerthe task of specifying it.For example, N itself is an N -module under usual addition. It represents theidentity functor on Mod N in an evident (and unique) way. A subset P of an N -module M issaid to be a sub- N -module if it admits an N -module structure making the inclusion P → M a homomorphism. Because a map of N -modules is injective if and only ifit is a monomorphism, we will usually identify monomorphisms and submodules.The dual statement is false—there are nonsurjective epimorphisms, for instancethe usual inclusion N → Z . The category has all products: Q i ∈ I M i is theusual product set with identity ( . . . , , . . . ) and componentwise addition( . . . , m i , . . . ) + ( . . . , m ′ i , . . . ) := ( . . . , m i + m ′ i , . . . ) . It also has all coproducts: L i ∈ I M i is the sub- N -module of Q i ∈ I M i consisting ofthe vectors ( m i ) i ∈ I such that m i = 0 for all but finitely many i ∈ I .In particular, we can construct free objects: given any set S and any set map f : S → M , the morphism L s ∈ S N → M defined by ( n s ) s ∈ S P s ∈ S n s f ( s ) isthe unique extension of f to an N -module map L i ∈ S N → M .82 J. Borger
A sub-set E ⊆ M × M is said to be a Mod N -equivalence relation if it is both an equivalencerelation on M and a sub- N -module of M × M .Given any homomorphism f : M → M ′ of N -modules, the induced equivalencerelation M × M ′ M is clearly a Mod N -equivalence relation. Conversely, given any Mod N -equivalence relation E on M , the set M/E of equivalence classes has aunique N -module structure such that the projection M → M/E sending x to theequivalence class [ x ] of x is a homomorphism of N -modules. In other words, therules [ x ] + [ y ] = [ x + y ] and 0 = [0] give a well-defined N -module structure on M/E . Let R be a subset of M × M . The Mod N -equivalence relation E generated by R is the minimal Mod N -equivalence relation on M containing R . It exists because any intersection of submodules is a submoduleand any intersection of equivalence relations is an equivalence relation. It canbe constructed explicitly by taking the transitive closure of the sub- N -module of M × M generated by R , the transpose of R , and the diagonal. Note the contrastwith the theory of modules over a ring, where taking the transitive closure isunnecessary. This gives the theory of modules over N a dynamical feel which isabsent when over a ring.We can construct N -modules in terms of generators and relations by combiningthis with free construction above. Clearly, much of this works in much moregeneral categories, especially categories of algebras, as defined below. We will usesuch generalizations without comment. In the present case and others, we willoften write M/ ( . . . , m i = n i , . . . ) for M/E if R = { . . . , ( m i , n i ) , . . . } . ⊗ . The set Hom(
M, P ) =
Mod N ( M, P ) of N -module homomor-phisms M → P is itself an N -module under pointwise addition:( f + g )( x ) := f ( x ) + P g ( x ) , x ) := 0 P . (1.6.1)We will also use the notation Hom N ( M, P ). It is functorial in M (contravariant)and P (covariant). For any fixed N -module M , the functor Hom( M, − ) has a leftadjoint, which we write M ⊗ − , or M ⊗ N − for clarity. In other words, M ⊗ M ′ ischaracterized by the property that a homomorphism M ⊗ M ′ → P is the same asset map h− , −i : M × M ′ → P which is N -bilinear in that it is an N -module mapin each argument if the other argument is fixed. It follows that if we denote theimage of an element ( m, m ′ ) under the universal bilinear map M × M ′ → M ⊗ M ′ by m ⊗ m ′ , then M ⊗ M ′ is the commutative monoid generated by symbols m ⊗ m ′ ,for all m ∈ M and m ′ ∈ M ′ , modulo all relations of the form( m + M m ) ⊗ m ′ = m ⊗ m ′ + m ⊗ m ′ , M ⊗ m ′ = 0 ,m ⊗ ( m ′ + M ′ m ′ ) = m ⊗ m ′ + m ⊗ m ′ , m ⊗ M ′ = 0 . Then ⊗ makes Mod N into a symmetric monoidal category with identity N . itt vectors, semirings, and total positivity N -algebras. An N -algebra (soon to be understood to be commutative) isdefined to be a monoid in Mod N with respect to the monoidal structure ⊗ . Thusan N -algebra is a set A with two associative binary operations + , × and respectiveidentity elements 0 , × distributes over + andsatisfies 0 × x = x × xy = x × y . We will sometimes usethe term semiring as a synonym for N -algebra.A morphism A → B of N -algebras is a morphism of monoids in the monoidalcategory Mod N . In other words, it is a function f : A → B which satisfies theidentities f (0) = 0 , f ( x + y ) = f ( x ) + f ( y ) , f (1) = 1 , f ( xy ) = f ( x ) f ( y ) . The category formed by N -algebras and their morphisms is denoted Alg N .For example, the N -module N admits a unique N -algebra structure; multipli-cation is usual integer multiplication. It is the initial object in Alg N . Likewise, 0with its unique N -algebra structure is the terminal object. For any subring A ⊆ R ,the subset A + := { x ∈ A | x ≥ } is a sub- N -algebra of A .The category of rings is equivalent in an evident way to the full subcategory of Alg N spanned by objects in which 1 has an additive inverse. From now on in this paper, all N -algebraswill be understood to be commutative under × unless stated otherwise. Howeverfor much of the rest of this section, this is just for convenience.Also for the rest of this section, A will denote an N -algebra. A -modules and A -algebras. One defines A -modules and A -algebras in theobvious way. An A -module (also called an A -semimodule in the literature) is an N -module equipped with an action of A with respect to the monoidal structure ⊗ .So it is an N -module M equipped with an N -module map A ⊗ M → M , written a ⊗ m am , such that the following identities are satisfied:1 m = m, ( ab ) m = a ( bm ) . A morphism of A -modules M → P is an N -linear map f : M → P satisfying theidentity f ( am ) = af ( m ). The category of A -modules is denoted Mod A . We willsometimes write Hom A ( M, P ) =
Mod A ( M, P ) for the set of A -module morphisms.Observe when A = N , the category just defined agrees with that defined in (1.1).An A -algebra is an N -algebra B equipped with a morphism i B : A → B . A morphism B → C of A -algebras is a morphism f : B → C of N -algebras such that f ◦ i B = i C . The category of A -algebras is denoted Alg A . As with modules, when A = N , the category of N -algebras as defined here agrees with that defined in (1.7).Also observe that if A is a ring, then these definitions of A -module and A -algebra agree with the usual ones in commutative algebra. In particular, a Z -module is the same as an abelian group, and a Z -algebra is the same as a ring.84 J. Borger A and ⊗ A . The set Hom A ( M, P ) has a natural A -module struc-ture given by pointwise operations. In other words, it is a sub- N -module ofHom N ( M, P ), and its A -module structure is given by the identity ( af )( x ) = af ( x ).Of course, this uses the commutativity of multiplication on A .For any fixed A -module M , the functor Hom A ( M, − ) has a left adjoint, whichwe write M ⊗ A − :Hom A ( M ⊗ A M ′ , N ) = Hom A ( M ′ , Hom A ( M, N )) . (Again, when A = N this agrees with the functor M ⊗ − defined above.) As above,an A -linear map M ⊗ A M ′ → P is the same a set map h− , −i : M × M ′ → P whichis A -bilinear in the sense that it is an A -module map in each argument when theother argument is held fixed. Thus M ⊗ A M ′ equals the quotient of M ⊗ N M ′ byall relations of the form ( am ) ⊗ m ′ = m ⊗ ( am ′ ) , (1.10.1)with its A -module structure given by a ( m ⊗ m ′ ) = am ⊗ m ′ = m ⊗ am ′ . A -modules and A -algebras. The category
Mod A has all limits and colimits. Limits, coproducts, and filtered colimits canbe constructed as when A is a ring, but coequalizers might be less familiar.Given a pair of maps f, g : M → P in Mod A , the coequalizer can be constructedas the quotient of P by the Mod A -equivalence relation generated by the subset { ( f ( x ) , g ( x )) | x ∈ M } ⊆ P × P . There are reasonable notions of kerneland cokernel, but we will not need them. The kernel of a map f : M → N of A -modules is defined to be the pull-back M × N (0) → M , and the cokernel is thepush-out N → N ⊕ M (0). Many familiar properties of kernels and cokernels fromabelian categories fail to hold for modules over semirings. For instance the summap N ⊕ N → N has trivial kernel and cokernel, but it is not an isomorphism.So kernels and cokernels play a less prominent role here than they do in abeliancategories. Equalizers and coequalizers are more useful. Let B be an A -algebra, and let M be an A -module. Then B ⊗ A M and Hom A ( B, M ) are B -modules in the evident ways. These constructions give the left and right adjointsof the forgetful functor U from B -modules to A -modules. They are called the B -modules induced and coinduced by A . It is also clear that the forgetful functor U is both monadic and comonadic. A -algebras. Like
Mod A , the category Alg A alsohas all limits and colimits. Limits, coproducts, and filtered colimits can again beconstructed as when A is a ring. In particular, coproducts are tensor products: B ∐ C = B ⊗ A C. itt vectors, semirings, and total positivity coproduct of any family ( B i ) i ∈ I is the tensor product of all B i over A . And as with modules, the coequalizer of two A -algebra morphisms f, g : B ⇒ C is C/R , where R is the equivalence relation internal to Alg A on C generated by the relation { ( f ( x ) , g ( x )) ∈ C × C | x ∈ B } . For any A -algebra B , the forgetful functor Alg B → Alg A has a left adjoint. It sends C to B ⊗ A C , where the B -algebrastructure is given by the map b b ⊗ Let M be an A -module. Because the func-tor M ⊗ A − : Mod A → Mod A has a right adjoint, it preserves all colimits. Sincefinite products and coproducts agree, it also preserves finite products. If it pre-serves equalizers, we say M is flat . In this case, M ⊗ A − preserves all finite limits.Observe that while not all monomorphisms are equalizers, it is nevertheless truethat tensoring with a flat module preserves monomorphisms. Indeed f : N → P isa monomorphism if and only if the diagonal map N → N × P N is an isomorphism,and this property is preserved by tensoring with a flat module.Flatness is preserved under base change. An A -algebra is said to be flat if itis flat when regarded as an A -module. If A → B and B → C are flat, then so isthe composition A → C . More generally, if B is a flat A -algebra, and M is a flat B -module, then M is flat as an A -module. Any free module is flat. Any filtered colimitof flat modules is flat. We will see in (2.5) below that flatness is a flat-local property.So for example a module is flat if it is flat-locally free.It is a theorem of Govorov and Lazard that any flat module over a ring can berepresented as a filtered colimit of free modules. This continues to hold for modulesover any N -algebra. As over rings, this is tantamount to an equational criterionfor flatness for modules over N -algebras, but now we must consider all relations ofthe form P i a i x i = P i b i x i instead of just those of the form P i a i x i = 0, as oneusually does over rings. See Katsov [22].If S is a multiplicative subset of A , let A [1 /S ] denote the initial A -algebrain which every element of S becomes multiplicatively invertible. Then A [1 /S ] isflat because it can be represented as colim s ∈ S A , where for all s, t ∈ S there is atransition map A → A from position s to position st given by multiplication by t .But it is completely different if we adjoin additive rather than multiplicativeinverses. We will see in (2.12) below that Z is not a flat N -module. In fact, 0 isthe only Z -module that is flat over N . It is also the only R -vector space that is flatover R + .
2. The flat topology over N The purpose of this section is to give some idea of scheme theory over N . It is thepoint of view I prefer for the mathematics of this chapter, but I will not use it ina serious way.86 J. Borger
Scheme theory and the flat topology over N were apparently first considered inTo¨en–Vaqui´e [44]. Lorscheid has considered a different but related approach [31](or see his chapter in this volume). In recent years, positivity structures in algebraicgeometry have appeared in some interesting applications, although in an ad hoc way. For example, let us mention the work of Lusztig [32], Fock–Goncharov [15],and Rietsch [36]. Let us say that a family ( B i ) i ∈ I of flat A -algebras is faithful if the family of base change functors Mod A −→ Y i ∈ I Mod B i , M ( B i ⊗ A M ) i ∈ I (2.1.1)reflects isomorphisms, that is, a map M → N of A -modules is an isomorphism if(and only if) for every i ∈ I the induced map B i ⊗ A M → B i ⊗ A N is. Let us saythat a family of flat A -algebras is an fpqc cover if it has a finite subfamily that isfaithful. For any N -algebra K , let Aff K denote the oppositeof the category of K -algebras. For any K -algebra A , write Spec ( A ) for the cor-responding object in Aff K . The fpqc covers form a pretopology on Aff K , in theusual way. See To¨en–Vaqui´e, proposition 2.4 [44]. The resulting topology is calledthe fpqc topology or, less formally, the flat topology .One might also like to define an fppf topology topology by requiring that each B i be finitely presented as an A -algebra. The following question is then natural: Question 1.
Let ( B i ) i ∈ I be a faithful family of flat A -algebras. If each B i isfinitely presented as an A -algebra, is there a finite faithful subfamily? When K is Z , it is a fundamental fact from scheme theory that the answer isyes. To prove it one combines quasi-compactness in the Zariski topology with thefact that faithfully flat morphisms of finite presentation have open image. Let ( B i ) i ∈ I be a faithful family of flat A -algebras.Then the family of base change functors (2.1.1) is comonadic. As usual, this is justan application of Beck’s theorem in category theory. (See Borceux [2], theorem4.4.4, p. 212. See also theorem 2.5 of To¨en–Vaqui´e [44].) Thus the fibered categoryof modules satisfies effective descent in the comonadic sense. If the family ( B i ) i ∈ I is finite, or more generally an fpqc cover, then the comonadic approach to descentagrees with the Grothendieck’s original one. So in either sense, in the fpqc topology,the fibered category of modules satisfies effective descent, or it is a stack.Thus descent allows us to recover the category of A -modules from that ofmodules over the cover. As usual, this allows us to recover A itself: Let A be an N -module, and let E denote the N -algebra (possiblynoncommutative, a priori) of natural endomorphisms of the identity functor on Mod A . Then the canonical map A → E is an isomorphism. itt vectors, semirings, and total positivity Proof.
Let ϕ be such a natural endomorphism. Set a = ϕ A (1). Then for any M ∈ Mod A , the map ϕ M : M → M is multiplication by a . Indeed, for any m ∈ M ,consider the map f : A → M determined by f (1) = m . Then we have ϕ M ( m ) = ϕ M ( f (1)) = f ( ϕ A (1)) = f ( a ) = am . Let B be an N -algebra, and let ( C i ) i ∈ I be a faithful family offlat B -algebras. (1) For any finite diagram ( M j ) j ∈ J of B -modules, a map M → lim j M j is anisomorphism if and only if each map C i ⊗ B M → lim j C i ⊗ B M j is anisomorphism. (2) Suppose B is an A -algebra, for some given N -algebra A . Then B is flat over A if and only if each C i is. (3) A B -module N is flat if and only if each C i ⊗ B N is flat as a C i -module.Proof. (1): First, because the family ( C i ) i ∈ I is faithful, the map M → lim j M j isan isomorphism if and only if each map C i ⊗ B M → C i ⊗ B lim j M j is. Second,because each C i is flat, we have C i ⊗ B lim j M j = lim j ( C i ⊗ B M j ). Combiningthese two statements proves (1).(2): Consider a finite diagram ( M j ) j ∈ J of A -modules. Suppose each C i is flatover A . Then the induced maps C i ⊗ A lim j M j → lim j ( C i ⊗ A M j ) are isomorphisms,and hence so are the maps C i ⊗ B B ⊗ A lim j M j → lim j ( C i ⊗ B B ⊗ A M j ) . Therefore by part (1), the map B ⊗ A lim j M j → lim j ( B ⊗ A M j ) is an isomorphism,and hence B is flat over A .The converse holds because flatness is stable under composition.(3): Suppose each C i ⊗ B N is flat over C i , and hence over B . Then for anyfinite diagram ( M j ) j ∈ J of B -modules, the maps C i ⊗ B N ⊗ B lim j M j → lim j ( C i ⊗ B N ⊗ B M j )are isomorphisms. Therefore by part (1), the map N ⊗ B lim j M j → lim j ( N ⊗ B M j )is an isomorphism, and so N is flat over B .The converse holds because flatness is stable under base change. N . We can then define the basic objects ofalgebraic geometry over any N -algebra K in a formal way, as in To¨en–Vaqui´e [44].A map f : Spec ( B ) → Spec ( A ) is Zariski open if the corresponding map A → B isa flat epimorphism of finite presentation. One then defines K -schemes by gluingtogether affine K -schemes along Zariski open maps. In To¨en–Vaqui´e, all this takesplace in the category of sheaves of sets on Aff K in the Zariski topology.Presumably one could define a category of algebraic spaces over N by adjoiningquotients of fppf or ´etale equivalence relations, under some meaning of these terms,88 J. Borger as for example in To¨en–Vaqui´e, section 2.2 [43]. But there are subtleties whetherone uses fppf maps, as in the question above, or ´etale maps, where it is not clearthat the first definition that comes to mind is the best one. So some care seems tobe needed before we can have complete confidence in the definition. In any case,we will certainly not need this generality.
Because we havefaithfully flat descent for modules, we can define the category
Mod X of X -modules(i.e., quasi-coherent sheaves) over any K -scheme X . More generally, any fiberedcategory over Aff K for which we have effective descent extends uniquely (up to someappropriate notion of equivalence) to such a fibered category over the category of K -schemes. For example, flat modules have this property by (2.5)(3). So we canmake sense of a flat module over any N -scheme.Similarly, using (2.5)(2), there is a unique way of defining flatness for morphisms X → Y of K -schemes that is stable under base change on Y and fpqc-local on X .We will conclude this section with some examples of flat-local constructionsand properties. An element m of an A -module M is additively idempotent if 2 m = m . The set I ( M ) of such elements is therefore theequalizer I ( M ) / / M x x / / x x / / M. Thus I has a flat-local nature. Indeed, for any flat A -algebra B , the induced map B ⊗ A I ( M ) → I ( B ⊗ A M ) is an isomorphism of B -modules, and if ( B i ) i ∈ J is anfpqc cover of A , then the induced map I ( M ) / / Y i I ( B i ⊗ A M ) / / / / Y j,j ′ I ( B j ⊗ A B j ′ ⊗ A M )is an equalizer diagram. It follows that, given an N -scheme X , this defines an X -module I ( M ) for any X -module M . Thus the functor I prolongs to a morphismof fibered categories I : Mod X → Mod X . An A -module M is additively cancellative if theimplication x + y = x + z ⇒ y = z holds in M . This is equivalent to the following being an equalizer diagram: M x,y ) ( x,y,y ) / / M x,y,z ) x + y / / ( x,y,z ) x + z / / M. Therefore, by (2.5), additive cancellativity is a flat-local property. itt vectors, semirings, and total positivity a ∈ A , let us say that M is a -cancellative if the implication ax = ay ⇒ x = y holds in M . This is equivalent to the following being an equalizer diagram: M x ( x,x ) / / M x,y ) ax / / ( x,y ) ay / / M Then a -cancellativity is also a flat-local property. A morphism M → N of A -modulesinduces two diagrams: M × M + / / (cid:15) (cid:15) M (cid:15) (cid:15) M × M + / / (cid:15) (cid:15) M (cid:15) (cid:15) N × N + / / N M × N + / / N. We say it is strong if the first diagram is Cartesian, and subtractive if the secondis. So a submodule M ⊆ N is subtractive if and only if it closed under differencesthat exist in N . Both properties are flat-local. For any N -algebra A and any A -module M , consider the subset of additively invertible elements: V ( M ) := { x ∈ M | ∃ y ∈ M, x + y = 0 } . (2.11.1)Then V ( M ) is a sub- A -module of M . The resulting functor is the right adjoint ofthe forgetful functor Mod Z ⊗ N A → Mod A , and so we have V ( M ) = Hom A ( Z ⊗ N A, M ) . It can also be expressed as an equalizer: V ( M ) x ( x, − x ) / / M / / + / / M (2.11.2)Therefore V has a local nature, and so we can define an X -module V ( M ) forany module M over any N -scheme X . In fact, since V ( M ) is a group underaddition, V ( M ) is an ( X × Spec ( Z ))-module, and so V can be viewed as a morphism Mod X → f ∗ ( Mod X × Spec ( Z ) ) of fibered categories, where f denotes the canonicalprojection X × Spec ( Z ) → X .Golan [17] says M is zerosumfree if V ( M ) = 0, or equivalently if (0) is astrong submodule. The remarks above imply that being zerosumfree is a flat-localproperty.90 J. Borger
Let A be a zerosumfree N -algebra, and let M be a flat A -algebra.Then M is zerosumfree. In particular, the zero module (0) is the only flat A -modulewhich is a group under addition, and the map A → Z ⊗ N A is flat if and only if Z ⊗ N A = 0 .Proof. Since M is flat, we have V ( M ) = M ⊗ A V ( A ) = M ⊗ A Z -scheme that is flat over a given zerosumfreescheme is the empty scheme. Thus we would expect some subtleties in the spiritof derived functors when passing from algebraic geometry over N to that over Z ,or from R + to R .
3. Plethystic algebra for N -algebras Let
K, K ′ , L be N -algebras. C objects in Alg K . Let C be a category of the kind consid-ered in universal algebra. Thus an object of C is a set with a family of multinaryoperations satisfying some universal identities. For example, C could be the cat-egories of groups, monoids, L -algebras, L -modules, Lie algebras over L , loops,heaps, and so on.Let us say that a covariant functor Alg K → C is representable if its underlyingset-valued functor is. Let us call the object representing such a functor a co- C object in Alg K . For example, a co-group structure on a K -algebra A is thesame as a group scheme structure on Spec ( A ) over Spec ( K ). Likewise, a co- L -algebra structure on A is the same as a L -algebra scheme structure on Spec ( A )over Spec ( K ). L -algebra objects in Alg K . Unpacking this further, we see a co- L -algebra object of Alg K is a K -algebra P together with K -algebra maps∆ + : P −→ P ⊗ K P ∆ × : P −→ P ⊗ K P (3.2.1)subject to the condition that for all A ∈ Alg K , the set Hom K ( P, A ) equipped withthe binary operations + and × induced by ∆ + and ∆ × is an N -algebra, plus an N -algebra map β : L −→ Alg K ( P, K ) . (3.2.2)These properties can of course be expressed in terms of P itself, without quantifyingover any variable algebras, as above with A . For example, ∆ + and ∆ × must becocommutative and coassociative, and ∆ × should codistribute over ∆ + , and so on.Similarly, the (unique) elements 0 and 1 in the N -algebras Hom K ( P, A ) corre-spond to K -algebra morphisms ε + : P → K, ε × : P → K. (3.2.3) itt vectors, semirings, and total positivity K - L -bialgebra instead of co- L -algebra object of Alg K .This is not meant to suggest any relation to the usual meaning of the term bialgebra in the theory of Hopf algebras. (Every K - L -bialgebra has two bialgebra structuresin the usual sense—∆ + and ∆ × —but this is just a coincidence of terminology.)Let Alg
K,L denote the category of K - L -bialgebras. A morphism P → P ′ of Alg
K,L is a K -algebra map compatible with the co-operations ∆ + , ∆ × , and β .In other words, the induced natural transformation of set-valued functors mustprolong to a natural transformation of L -algebra-valued functors. In the case where K and L are Z -algebras, the the-ory of K - L -bialgebras was initiated in Tall–Wraith [41] and developed further inBorger–Wieland [8]. It is clear how to extend the general theory developed thereto N -algebras. In almost all cases, the relevant words from [8] work as written;at some places, obvious changes are needed. The reader can also refer to Stacey–Whitehouse [39], where the general theory is written down for general universal-algebraic categories. (Also, see Bergman–Hausknecht [1] for many fascinating casestudies taken from different categories, such as Lie algebras, monoids, groups, pos-sibly noncommutative rings, and many more.)Let us list some of the main ideas we will need.(1) The composition product is a functor − ⊙ L − : Alg
K,L × Alg L → Alg K . It ischaracterized by the adjunction Alg K ( P ⊙ L A, B ) =
Alg L ( A, Alg K ( P, B )) . (2) It has an extension to a functor − ⊙ L − : Alg
K,L × Alg
L,K ′ → Alg
K,K ′ (3) This gives a monoidal structure (not generally symmetric) on the category Alg
K,K of K - K -bialgebras. The unit object is K [ e ], the one representing theidentity functor.(4) A composition K -algebra is defined to be a monoid with respect to thismonoidal structure. The operation is denoted ◦ and the identity is denoted e .(5) An action of a composition K -algebra P on a K -algebra A is defined to bean action of the monoid object P , or equivalently of the monad P ⊙ K − . Wewill write f ( a ) for the image of f ⊙ a under the action map P ⊙ K A → A . A P -equivariant K -algebra is a K -algebra equipped with an action of P . When K = N , we will also use the term P -semiring .(6) For any K - L -bialgebra P , we will call the functor it represents its Witt vector functor: W P = Alg K ( P, − ). It takes K -algebras to L -algebras. When K = L , It is called a biring triple in [41], a plethory in [8], a Tall-Wraith monad object in [1], anda Tall-Wraith monoid in [39]. The term composition algebra is both plain and descriptive; so Ithought to try it out here. It does however have the drawback in that it already exists in theliterature with other meanings. J. Borger a composition structure on P is then equivalent to a comonad structure on W P . When P has a composition structure, then W P ( A ) has a natural actionof P , and in this way W P can be viewed as the right adjoint of the forgetfulfunctor from P -equivariant K -algebras to K -algebras. An element ψ of a composition K -algebra P is K -algebra-like if for all K -algebras A with anaction of P , the self map x ψ ( x ) of A is a K -algebra map. This is equivalent torequiring∆ + ( ψ ) = ψ ⊗ ⊗ ψ, ∆ × ( ψ ) = ψ ⊗ ψ, β ( c )( ψ ) = c, (3.4.1)for all c ∈ K . (For comparison, one could say d ∈ P is K -derivation-like if it actsas a K -linear derivation on any K -algebra. This can also be expressed directly bysaying d is primitive under ∆ + and it satisfies the Leibniz rule ∆ × ( d ) = d ⊗ e + e ⊗ d and the K -linearity identity β ( c )( d ) = 0.)Now let G be a monoid. Let P be the K -algebra freely generated (as an algebra)by the symbols ψ g , for all g ∈ G . Then P has a unique composition structure suchthat each ψ g is K -algebra-like and we have ψ g ◦ ψ h = ψ gh for all g, h ∈ G . Then an action of P on an algebra A , in the sense of (5) of3.3, is the same as an action of G on A , in the usual sense of a monoid map G → Alg K ( A, A ).In this case, the Witt functor is simply W P ( A ) = A G , where A G has the usualproduct algebra structure. C objects in Alg K . Let C be a category of the kind con-sidered above. Let K → K ′ be an N -algebra map, and let P ′ be a co- C object in Alg K ′ . Then a model for P ′ over K (or a K -model ) is a co- C object P in Alg K together with an isomorphism K ′ ⊗ K P → P ′ of co- C objects of Alg K ′ . Then forany K ′ -algebra A ′ , we have W P ′ ( A ′ ) = Alg K ′ ( P ′ , A ′ ) ∼ −→ Alg K ( P, A ′ ) = W P ( A ′ ) . So the Witt vector functor of P extends W P ′ from Alg K ′ to Alg K . Conversely, anysuch extension to a representable functor comes from a unique model of P ′ . C -objects in Alg K . We will be especially interestedin finding K -models of P ′ that are flat (over K ). Of course, these can exist onlywhen P ′ is flat over K ′ , but this will be the case in all our examples. Further wewill only consider the case where the structure map K → K ′ is injective.Under these assumptions, the composition P → K ′ ⊗ K P → P ′ is injective.Conversely, if a subset P ⊂ P ′ admits a flat model structure, then it does so in aunique way. Indeed, since the induced maps P ⊗ K n → P ′⊗ K ′ n are injective, eachco-operation ∆ on P ′ restricts to at most one on P . One might then say that being itt vectors, semirings, and total positivity K ⊆ K ′ ) is a property of a given subset of P ′ , rather than astructure on it.The case where C = Alg L will be of particular interest to us. Then a flat modelis just a subset P ⊆ P ′ such that the following properties hold:(1) P is a flat sub- K -algebra of P ′ ,(2) the induced map K ′ ⊗ K P → P ′ is a bijection,(3) ∆ + ( P ) ⊆ P ⊗ K P and ε + ( P ) ⊆ K ,(4) ∆ × ( P ) ⊆ P ⊗ K P and ε × ( P ) ⊆ K ,(5) β ( L ) ⊆ Alg K ( P, K ), where
Alg K ( P, K ) is regarded as a subset of
Alg K ′ ( P ′ , K ′ ).When L = N , the last condition is always satisfied. (Alternatively, the conditionson the co-units ε + and ε × are also redundant but not in the absence of (5).)For other categories C , it is usually clear how to modify these conditions. Forexample, if C = Mod N , one would drop (4) and (5). When P ′ is a composition K ′ -algebra,we will usually want to descend the composition structure as well. Then a K -modelof P ′ (as a composition algebra) is a composition K -algebra P together with anaction of P on K ′ and an isomorphism K ′ ⊗ K P → P ′ of composition K ′ -algebras.Giving such a model is equivalent to extending W P ′ to a representable comonadon Alg K .A flat model (when K ⊆ K ′ ) is then just a flat sub- K -algebra P ⊆ P ′ satisfyingconditions (1)–(5) above plus(6) P ◦ P ⊆ P and e ∈ P .So again, if P admits such a structure, it does so in a unique way.
4. The composition structure on symmetric functions over N The purpose of this section is to give two different N -models of Λ Z , the compositionring of symmetric functions. Since Λ Z represents the usual big Witt vector functor,these give extensions of the big Witt vector functor to N -algebras.Our treatment is broadly similar to Macdonald’s [33]. He discusses the co-additive structure in example 25 of I.5, the co-multiplicative structure in example20 of I.7, and plethysm in I.8.Let K be an N -algebra. We will follow those of Macdonald, p. 1 [33].So, a partition is an element λ = ( λ , λ , . . . ) ∈ N ⊕ N ⊕ · · · J. Borger such that λ ≥ λ ≥ · · · . As is customary, we will allow ourselves to omit any num-ber of zeros, brackets, and commas and to use exponents to represent repetition.So for example we have (3 , , , , , . . . ) = 32
10 = 32 , . . . ) = 0.The length of λ is the smallest i ≥ λ i +1 = 0. The weight of λ is P i λ i and is denoted | λ | . We also say λ is a partition of its weight. K . Let Ψ K denote the composition K -algebra associated to the multi-plicative monoid of positive integers. So Ψ K = K [ ψ , ψ , . . . ], where each ψ n is K -algebra-like and we have ψ m ◦ ψ n = ψ mn .We will be interested in (flat) models of Ψ Q over smaller subalgebras, especially Q + , Z and N . There are the obvious models Ψ Q + , Ψ Z , and Ψ N , but we will bemore interested in larger ones. Let Λ K denote the usual K -algebra of symmetricfunctions in infinitely many variables x , x , . . . with coefficients in K . (See Mac-donald, p. 19 [33].) More precisely, Λ K is the set of formal series f ( x , x , . . . ) suchthat the terms of f have bounded degree and for all n , the series f ( x , . . . , x n , , , . . . )is a polynomial which is invariant under permuting the variables x , . . . , x n .It is clear that Λ K is freely generated as a K -module by the monomial sym-metric functions m λ , where λ ranges over all partitions and where m λ = X α x α x α · · · where α runs over all permutations of λ in N ∞ . In particular, we have K ⊗ N Λ N =Λ K . When K = N , this is the unique basis of Λ N , up to unique isomorphism onthe index set.It is well known that when K is a ring, Λ K is freely generated as a K -algebraby the complete symmetric functions h , h , . . . , where h n := X i ≤···≤ i n x i · · · x i n = X | λ | = n m λ . Alternatively, if we write ψ n = x n + x n + · · · , then the induced map Ψ K = K [ ψ , ψ , . . . ] −→ Λ K (4.3.1)is an injection when K is a flat Z -algebra and is a bijection when K is a Q -algebra.In particular, Λ N and Λ Z are models for Ψ Q = Λ Q . The elements ψ n have severalnames: the Adams , Frobenius , and power-sum symmetric functions. N is not free as an N -algebra. Indeed, all m λ are indecompos-able additively and, one checks, multiplicatively—except m , which is invertible.Therefore any generating set of Λ N as an N -algebra must contain all the m λ but m . But they are not algebraically independent, because any monomial in themis a linear combination of the others. For example m = m + 2 m , . itt vectors, semirings, and total positivity At times, we will use otherfamilies of symmetric functions, such as the elementary symmetric functions e n = X i < ···
1. Probably the most concise way of relating them all is with generatingfunctions in 1 + t Λ Q [[ t ]]: Y d ≥ (cid:0) − θ d t d (cid:1) − = exp (cid:0) X j ≥ ψ j j t j (cid:1) = X i ≥ h i t i = (cid:0) X i ≥ e i ( − t ) i (cid:1) − . (4.5.2)Indeed, one can check that each expression equals Q j (1 − x j t ) − .One notable consequence of (4.5.2) is that the θ d generate Λ Z freely as a ring:Λ Z = Z [ θ , θ , . . . ] . Another is that we have Q + [ h , h , . . . ] ⊆ Ψ Q + . (4.5.3)Note that such a containment is a special property of the complete symmetricfunctions h n ; it is not shared by the elementary symmetric functions e n .The Witt symmetric functions are rarely encountered outside the literature on96 J. Borger
Witt vectors. The following are some decompositions in more common bases: θ = ψ = m = e = h = s θ = ( − ψ + ψ ) / − m = − e = − h + h = − s θ = ( − ψ + ψ ) / − m − m = − e e + e = − h h + h = − s θ = ( − ψ + 2 ψ ψ − ψ + 2 ψ ) / − m − m − m − m = − e e + e e − e = − h + 2 h h − h − h h + h = − s − s − s − s θ = ( − ψ + ψ ) / − m − m − m − m − m − m = − e e + e e + e e − e e − e e + e = − h h + h h + h h − h h − h h + h = − s − s − s − s − s θ = ( − ψ − ψ ψ + 9 ψ ψ − ψ + 8 ψ ψ − ψ + 12 ψ ) / − m − m − m − m − m − m − m − m − m − m = − e e + e e + e e − e e e − e e + e e + e e − e = − h h − h h + 3 h h h − h + h h − h h − h h + h = − s − s − s − s − s − s − s − s Observe that the coefficients in some bases are noticeably smaller than in others.It would be interesting to make this precise.
Consider the composition algebra structure on Λ Q induced bythe isomorphism Ψ Q → Λ Q of (4.3.1). Then the structure maps on Λ Q satisfy thefollowing: ∆ + : f ( . . . , x i , . . . ) f ( . . . , x i ⊗ , ⊗ x i , . . . )∆ × : f ( . . . , x i , . . . ) f ( . . . , x i ⊗ x j , . . . ) ε + : f ( . . . , x i , . . . ) f (0 , , . . . ) ε × : f ( . . . , x i , . . . ) f (1 , , , . . . ) f ◦ g = f ( y , y , . . . ) , for all f ∈ Λ Q and g ∈ Λ N , where g = y + y + · · · with each y j a monomial inthe variables x i with coefficient .Proof. The maps ∆ + , ∆ × , ε + , ε × displayed above are manifestly Q -algebra maps.Therefore to show they agree with the corresponding structure maps, it is enoughto consider elements f that run over a set of generators, such as the ψ n . In thiscase, one can calculate the image of ψ n under the maps displayed above and observethat it agrees with the image of ψ n under the structure maps, by (3.4.1).Similarly, for any fixed g ∈ Λ N , the map f f ◦ g displayed above is a Q -algebra map. So it is again enough to assume f = ψ m . Then ψ m ◦ g g is an itt vectors, semirings, and total positivity N -algebra map Λ N → Λ Q , and to show this map agrees with the rule displayedabove, it is enough to do so after allowing coefficients in Q , since Λ N is free. Thenthe ψ n are generators, and so it is enough to assume g = ψ n . We are left to check ψ m ◦ ψ n = ψ mn , which is indeed the composition law on Ψ Q . Λ N and Λ Z are models of the composition algebra Λ Q in a uniqueway.Proof. Since both Λ N and Λ Z are free, it is enough to check properties (1)–(6)of (3.6) and (3.7). It is immediate from (4.6) that all the structure maps preserveΛ N , which finishes the proof for it.It is also clear that all the structure maps preserve Λ Z , with the exception ofthe one expressing that Λ Z is closed under composition. We will now show this.Because Λ N is a composition algebra and we have Λ Z = Z ⊗ N Λ N , it is enough toshow f ( Z ) ⊆ Z for all f ∈ Λ Z , and hence f ( − ∈ Z for all such f . To do this, itis enough to restrict to the case where f ranges over a set of algebra generators,such as the h n . But since ψ n ( −
1) = −
1, the identity (4.5.2) implies h n ( −
1) = 0for n ≥ h ( −
1) = −
1. So we have h n ( − ∈ Z for all n . N -action. For the convenience of the reader,let us spell out what it means for the composition algebra Λ N to act on an N -algebra A in elementary terms.For each partition λ , there is a set map m λ : A −→ A such that the following identities hold m ( x ) = 1( m λ m µ )( x ) = m λ ( x ) m µ ( x ) m λ (0) = ε + ( m λ ) m λ ( x + y ) = ∆ + ( m λ )( x, y ) := X µ,ν q λµν m µ ( x ) m ν ( y ) m λ (1) = ε × ( m λ ) m λ ( xy ) = ∆ × ( m λ )( x, y ) := X µ,ν r λµν m µ ( x ) m ν ( y ) m ( x ) = xm λ ( m µ ( x )) = ( m λ ◦ m µ )( x ) . The notation here requires some explanation. The expression ( m λ m µ )( x ), or moregenerally f ( x ), is defined by f ( x ) = X ν f ν m ν ( x ) ∈ A, J. Borger where the numbers f ν ∈ N are defined by the equation f = P ν f ν m ν ; and q λµν and r λµν are the structure coefficients in N for the two coproducts:∆ + ( m λ ) = X µ,ν q λµν m µ ⊗ m ν ∆ × ( m λ ) = X µ,ν r λµν m µ ⊗ m ν . N -structure on monoid algebras. Let A be acommutative monoid, written multiplicatively. The monoid algebra N [ A ] is the setof finite formal sums P ni =1 [ a i ], where each a i ∈ A , and multiplication is the linearmap satisfying the law [ a ][ b ] = [ ab ]. Then for any f ∈ Λ N define f ( n X i =1 [ a i ]) := f ([ a ] , . . . , [ a n ] , , , . . . ) . (4.9.1)The right-hand side denotes the substitution x i = [ a i ] into the symmetric function f . To show this law defines an action of Λ N on N [ A ], it is enough to observethat (4.9.1) is the restriction of an action of a larger composition algebra on a larger N -algebra. By (4.6), it is the restriction of the action of Λ Q on Q [ A ] := Q ⊗ N N [ A ]determined by ψ n ([ a ]) = [ a n ] for all a ∈ A , n ≥ N -structure on N [ A ] the toric Λ N -structure because Spec ( Z [ A ]) isa toric variety and the Λ N -structure extends in a canonical way to any nonaffinetoric variety, once this concept is defined. (See [3], for example.) When A = Z above, we have N [ A ] = N [ x ± ]. Let B denote the subalgebra of N [ A ] spanned by 1 , x + x − , x + x − , . . . .Then B is the set of invariants under the involution ψ − of N [ x ± ] defined by x x − . Since ψ − is a Λ N -morphism and since the category of Λ N -algebras hasall limits, B is a sub-Λ N -semiring. It is a model over N of the Λ Z -ring Z [ x + x − ]called the Chebyshev line in [3], but it is not isomorphic to the N -algebra N [ y ]. Infact, it is not even finitely generated as an N -algebra. N -semirings. In [6], De Smit and I proved some classifica-tion results about reduced Λ-rings that are finite flat over Z . It would be interestingto know if there are similar results over N . As D. Grinberg pointed out to me, it isnot hard to construct non-toric Λ N -semirings which are flat and finitely presentedover N . One example is the sub- N -algebra of ψ − -invariants of the monoid algebra N [ x ] / ( x = 1). It is isomorphic to N [ y ] / ( y = y + 2), via the map y x + x . Let B be asubmonoid of C under addition, and let N [ e B ] b denote the set of formal series P ∞ i =1 [ e b i ] , where b i ∈ B , such that the complex series P i e b i converges absolutely.(Of course, e now denotes the base of the natural logarithm.) We identify seriesthat are the same up to a permutation of the terms. More formally, N [ e B ] b is the itt vectors, semirings, and total positivity . . . , n b , . . . ) ∈ N B such that the sums P b ∈ S n b | e b | are bounded as S ranges over all finite subsets of B . (Also, note that the notation is slightly abu-sive in that N [ e B ] b depends on B itself and not just its image under the exponentialmap.) It is a sub- N -module of N B and has a multiplication defined by (cid:16) X i [ e b i ] (cid:17)(cid:16) X j [ e c j ] (cid:17) = (cid:16) X i,j [ e b i + c j ] (cid:17) , under which it becomes an N -algebra. It is the unique multiplication extendingthat on the monoid algebra N [ B ] which is continuous, in some suitable sense.For integers n ≥
1, define ψ n (cid:16) X i [ e b i ] (cid:17) = X i [ e nb i ] . This is easily seen to be an element of N [ e B ] b , and hence the ψ n form a commutingfamily of N -algebra endomorphisms of N [ e B ] b . The induced endomorphisms of Q ⊗ N N [ e B ] b prolong to a unique action of Λ Q . It follows that for any symmetricfunction f ∈ Λ Q , we have f (cid:0) X i [ a i ] (cid:1) = f ([ a ] , [ a ] , . . . ) ∈ Q ⊗ N N B . (4.12.1)Indeed, it is true when f = ψ n ; then by the multiplication law above, it is truewhen f is a monomial in the ψ n , and hence for any f ∈ Ψ Q = Λ Q . It follows thatfor any f ∈ Λ N , the element f ( P i [ a i ]) lies in N B ∩ ( Q ⊗ N N [ e B ] b ) = N [ e B ] b . So N [ e B ] b inherits an action of Λ N from Q ⊗ N N [ e B ] b .Observe that if B is closed under multiplication by any real t ≥
1, then wehave operators ψ t , for all t ≥ ψ t (cid:0) X i [ e b i ] (cid:1) = X i [ e tb i ] . Indeed the series P i e tb i is absolutely convergent if P i e b i is. Thus Spec ( A ) has aflow interpolating the Frobenius operators. For further remarks, see (8.11). Now let A be a submonoid ofthe positive real numbers under multiplication, and let log A = { b ∈ R | e b ∈ A } .Then define N [ A ] b = N [ e log A ] b . Z over N . The N -algebra N [ e , . . . ] is amodel for Λ Z as a co- N -module object because the expression∆ + ( e n ) = X i + j = n e i ⊗ e j J. Borger has no negative coefficients. But it is not a model as a co- N -algebra object becausethe analogous expression for ∆ × does at places have negative coefficients:∆ × ( e ) = e ⊗ e + e ⊗ e − e ⊗ e . The complete symmetric functions h n = X i ≤···≤ i n x i · · · x i n behave similarly. The N -algebra N [ h , . . . ] is a model for Λ Z over N as a co- N -module object, because ∆ + ( h n ) = P i + j = n h i ⊗ h j , but not as a co- N -algebraobject, because ∆ × ( h ) = h ⊗ h − h ⊗ h − h ⊗ h + 2 h ⊗ h .For Witt symmetric functions, we have∆ + ( θ p ) = θ p ⊗ ⊗ θ p − p − X i =1 p (cid:18) pi (cid:19) θ i ⊗ θ p − i ∆ × ( θ p ) = θ p ⊗ θ p + θ p ⊗ θ p + pθ p ⊗ θ p , for any prime p . So N [ θ , . . . ] is not even a model as a co- N -module object. Using − θ p instead of θ p , removes the sign from the first formula but adds one in thesecond. We will see in section 8 that it is possible to circumvent this problem, atleast if we care only about a single prime.
5. The Schur model for Λ Z over N Sch . For any partition λ = ( λ , . . . , λ r ), put s λ := det( h λ i − i + j ) ∈ Λ Z , (5.1.1)where as usual h = 1 and h n = 0 for n <
0. For example, s = det h h h h h h h . (5.1.2)Such symmetric functions are called Schur functions . They have simple interpre-tations in terms of representation theory of general linear and symmetric groups,and several of the results we use below are usually proved using such an interpre-tation. But it is enough for us just to cite the results, and so we will ignore thisinterpretation. This is discussed more in (5.11) below.Write Λ
Sch := M λ N s λ , (5.1.3)where λ runs over all partitions. The Schur functions are well known to form a Z -linear basis for Λ Z . (See p. 41 (3.3) of Macdonald’s book [33].) Therefore Λ Sch is an N -model for Λ Z as a module, by way of the evident inclusion Λ Sch → Λ Z . itt vectors, semirings, and total positivity Λ Sch is an N -model of Λ Z as a composition algebra, in a uniqueway. We also have Λ Sch ⊆ Λ N .Proof. Since Λ
Sch is free, it is enough to check properties (1)–(6) of (3.6) and (3.7).These all reduce to standard facts about Schur functions, for which we will referto chapter 1 of Macdonald’s book [33].We have Λ
Sch ⊆ Λ N because of the equality s λ = X µ K λµ m µ (5.2.1)where the K λµ are the Kostka numbers, which are in N , by Macdonald, p. 101(6.4).The fact that Λ Sch is a sub- N -algebra follows from the equations1 = s , s λ s µ = X ν c νλµ s ν , where the c νλµ are the Littlewood–Richardson coefficients, which are in N , by Mac-donald, pp. 113–114, (7.5) and (7.3).By Macdonald, p. 72 (5.9) and p. 119 (7.9), we have∆ + ( s λ ) = X µν c λµν s µ ⊗ s ν (5.2.2)∆ × ( s λ ) = X µν γ λµν s µ ⊗ s ν , (5.2.3)where the c νλµ are again the Littlewood–Richardson coefficients and the γ λµν arethe Kronecker coefficients, which are in N by pp. 114–115 of Macdonald. It followsthat Λ Sch is closed under the co-addition and comultiplication maps.The containments ε + (Λ Sch ) , ε × (Λ Sch ) ⊆ N follow from Λ Sch ⊆ Λ N .Finally, we have s λ ◦ s µ = X ν a νλµ s ν , where a νλµ ∈ N , by Macdonald, p. 136 (8.10). So we have s λ ◦ s µ ∈ Λ Sch . It followsthat Λ
Sch is closed under composition, since Λ
Sch is an N - N -bialgebra. Last, itcontains the compositional identity element e because e = s . The three models of Ψ Q over N we have seen are Ψ N , Λ N , andΛ Sch . The largest of the three is Λ N . The other two are incomparable, even over Q + . Indeed, we have s = ( ψ − ψ ) / ψ = s − s . Sch is not free as an N -algebra. As in (4.4), any generatingset of Λ
Sch contains all Schur functions that cannot be written as monomials withcoefficient 1 in the other Schur functions. This is true of s = h , s = h , and s = h − h because they are irreducible in Λ Z = Z [ h , h , . . . ]. But they are notalgebraically independent: s = s + s .02 J. Borger
Sch -action.
This is the same asin (4.8), but with operators s λ . Now the identities are the following: s ( x ) = 1( s λ s µ )( x ) = s λ ( x ) s µ ( x ) s λ (0) = ε + ( s λ ) s λ ( x + y ) = ∆ + ( s λ )( x, y ) := X µ,ν c λµν s µ ( x ) s ν ( y ) s λ (1) = ε × ( s λ ) s λ ( xy ) = ∆ × ( s λ )( x, y ) := X µ,ν γ λµν s µ ( x ) s ν ( y ) s ( x ) = xs λ ( s µ ( x )) = ( s λ ◦ s µ )( x ) := X ν a νλµ s µ ( x ) s ν ( y ) , where c λµν , γ λµν , and a λµν are the Littlewood–Richardson, Kronecker, and Schurplethysm coefficients, as in the proof of (5.2).In these terms, a Λ N -action gives rise to a Λ Sch -action by the formula (5.2.1).
Since we have Λ
Sch ⊆ Λ N , any Λ N -semiring is a Λ Sch -semiring.For example, in a monoid N -algebra N [ M ] with the toric Λ N -structure, we have s λ ([ m ]) = (cid:26) [ m ] l if λ = ( l )0 otherwise (5.6.1) Consider the monoid Z -algebra Z [ x ] / ( x −
1) with the Λ Z -structuredefined by ψ n ( x ) = x n , and let A denote the sub- N -algebra generated by η = − x .So we have a presentation A = N [ η ] / ( η = 1) . Then A is a sub-Λ Sch -semiring. Indeed, for any partition λ of r , we have s λ ( − x ) = ( − r s λ ′ ( x ) = (cid:26) ( − x ) r if λ = (1 r )0 otherwiseby (5.6.1), where λ ′ is the partition conjugate to λ . p -derivations. For each integer n ≥
1, we have ψ n = m n ∈ Λ N . Therefore every Λ N -semiring A has a natural endomorphism ψ n . It iscalled the Adams or Frobenius operator . When n is a prime p , the induced mapon F p ⊗ N A is the Frobenius map x x p . To see this, one can reduce to the caseof rings, or one can simply observe that we have d p := 1 p ( ψ p − ψ p ) ∈ Λ N itt vectors, semirings, and total positivity ψ p ( x ) + pd p ( x ) = x p . The operator − d p = θ p is a p -derivation in thelanguage of Buium, p. 31 [11].It is different for Λ Sch -semirings. Since ψ n Λ Sch for n ≥
2, we do not generallyhave Frobenius lifts. However we do have the operators d n := − θ n for n ≥ − θ n ∈ Λ Sch for n ≥ . Of course, for n = 1 we have the opposite: + θ ∈ Λ Sch . This irksome exceptionsuggests the perspective here might not be the most fundamental one. For ex-ample, it is a standard result in the theory of p -typical Witt vectors (tantamountto the Cartier–Dieudonn´e–Dwork lemma) that the ring Z [ d p n | n ≥
0] of p -typicalsymmetric functions agrees with Z [ d ◦ np | n ≥ Z -algebra gener-ated by the iterated p -derivations. The families of generators d p n and d ◦ np agreefor n ≤ n = 2. Question 2.
Do the symmetric functions d p n − d ◦ np lie in Λ Sch ? More generally,do d p m + n − d p m ◦ d p n ? If not, do they lie in Λ N ? If the answers are yes, it would suggest that the operators d ◦ np are more fun-damental from the point of view of positivity than the d p n . In other words, up tosome signs, the iterated p -derivations generate a larger sub- N -algebra than the Wittsymmetric functions. Computations have shown that the answer to the strongerquestion above is yes when p = 2, m + n ≤ p = 3, m + n = 2; and p = 5, m + n = 2.The obvious generalization d rs − d r ◦ d s ∈ Λ Sch is not always true. Computationshave shown it is true for ( r, s ) = (2 , , (2 , , (3 , , (5 ,
3) but false for (3 ,
2) and(5 , Observe that neitherΛ N nor Λ Sch is generated by linear operators, even after base change to R + . Indeed,because we have m = e = s , any generating set of Λ R + or R + ⊗ N Λ Sch wouldhave to contain a nonzero multiple of e , which is nonadditive element. So anaction of Λ Z or Λ Sch on an N -algebra cannot be expressed entirely in terms ofadditive operators.It appears to be the case that over Z or Z p , any composition algebra that cannotbe generated by linear operators can trace its origin to the d p operators—in otherwords, to lifting Frobenius maps to characteristic 0. There are probably theoremsto this effect. (For example, Buium [10] classifies ring scheme structures on theplane A Z p , which is perhaps the first test case, and he is able to prove such a resultthere.) Could it be that the existence of nonlinear composition algebras over R + isdue to a similarly identifiable phenomenon? It seems optimistic to hope that theanswer is yes, but if it were, the importance would be so great that the questionshould not be dismissed. The compositionring Λ Z has analogues over rings of integers in general number fields. We will not04 J. Borger really use them in this chapter, but they will appear in a several remarks and openquestions.Let K be a number field, let O K denote its ring of integers, and let E denotea family of maximal ideals of O K . Let Λ O K ,E denote the composition O K -algebracharacterized by the property that an action on any flat O K -algebra A is the sameas a commuting family of O K -algebra endomorphisms ( ψ p ) p ∈ E such that each ψ p reduces to the Frobenius map x x [ O K : p ] on A/ p A . This construction and theassociated Witt vector functor W O K ,E are discussed in much more detail in section1 of my paper [4]. When K = Q , it reduces to a special case of the constructionof (8.1), as is explained there, but otherwise there is no overlap between the two. The standard wayof looking at Λ
Sch is from the point of view of polynomial functors, or equiva-lently representations of GL n . (See Macdonald [33], ch. 1, app. A.) Let F denotethe category of polynomial functors from the category of finite-dimensional vectorspaces over, say, C to itself. Then for any polynomial functor F , the polynomialtr( F (diag( x , . . . , x n ))) in Z [ x , . . . , x n ] is symmetric, and as n varies, these poly-nomials define a compatible sequence and hence an element of the inverse limitΛ Z . This defines a group homomorphism χ : K ( F ) → Λ Z , where K ( F ) denotes theGrothendieck group of F , and one proves this map is bijection.Under this bijection the irreducible polynomial functors correspond to the Schurfunctions. Indeed, given a partition λ of n , let V λ be the corresponding irreducible Q -linear representation of S n . Then the functor S λ ( E ) := Hom Q [ S n ] ( V λ , E ⊗ n ) isa polynomial functor. Further, the S λ are precisely the irreducible polynomialfunctors, and each S λ corresponds to the Schur function s λ under χ . Therefore χ induces an isomorphism K + ( F ) → Λ Sch of N -modules, where K + ( − ) denotesthe subset of the Grothendieck group consisting of effective classes, rather thanjust virtual ones. Then each of the structure maps in the composition-algebrastructure on Λ Sch corresponds to something transparent on F : the operations +, × , and plethysm on Λ Sch correspond to ⊕ , ⊗ , and composition on F . The co-operation ∆ + corresponds to the rule that sends a polynomial functor F to thepolynomial functor F ( E ⊕ E ′ ) in two variables E and E ′ ; similarly ∆ + correspondsto F ( E ⊗ E ′ ). The positivity properties of Λ Sch used in the proof of (5.2) belowthen hold because the (co-)operations on Λ
Sch correspond to (co-)operations on F ,and hence they preserve effectivity.This point of view is also a good way of approaching the λ -ring structure onGrothendieck groups. Given an amenable linear tensor category C , we can definefunctors S λ : C → C by S λ ( E ) = Hom Q [ S n ] ( V λ , E ⊗ n ). This defines an action ofthe monoid K + ( F ) = Λ Sch under composition on K + ( C ), and this is essentially bydefinition an action of the composition N -algebra Λ Sch on the N -algebra K + ( C ).This, in turn, induces an action of Λ Z on K ( C ), which is the usual λ -ring structureon Grothendieck groups.While this is a more conceptual way of thinking about positivity propertieson Λ Sch , the deeper meaning of the connection with Witt vectors still eludes me.From an arithmetic point of view, it is not clear why one would should consider itt vectors, semirings, and total positivity Z . Indeed, thecomposition algebras Λ O K ,E of (5.10) and the corresponding Witt functors W O K ,E also have arithmetic interest (for instance in the theory of complex multiplicationwhen K is an imaginary quadratic field), but no representation theoretic interpre-tation of Λ O K ,E is known. In fact, no interpretation in terms of something similarto symmetric functions is known.At the time of this writing, there is still a tension between the following threeobservations: (1) the connection between Frobenius lifts and the representationtheory of GL n appears to be a coincidence, merely an instance of a fundamentalalgebraic object arising in two unrelated contexts; (2) to define Λ Sch and W Sch oneneeds the positivity results established with representation theory; (3) Λ
Sch and W Sch have some arithmetic interest, as in section 7 below. The situation with Λ N and W is perhaps less mysterious—they seem more important from the arithmeticpoint of view and also require no nontrivial positivity results from representationtheory. Perhaps the resolution will be that Λ Sch and W Sch are of arithmetic interestonly through their relation to Λ N and W .Whatever the case, this is why I have ignored K-theory and used the represen-tation theory as a black box. However this point of view does suggest the followingquestion: Question 3.
Let B p denote the subset of Λ Z consisting of the characters of poly-nomial functors over an infinite field of characteristic p > . Is B p a model for Λ Z as a composition algebra? The structure of B p is apparently much subtler than that of its analogue Λ Sch in characteristic 0. Nevertheless I expect the answer to this question to be yes for formal reasons, as with Λ
Sch above. It would be interesting to make a moredetailed study of the B p from the point of view of plethystic algebra.
6. Witt vectors of N -algebras For the classical theory of Witt vectors for rings, one can see Bergman’s lecture 26in Mumford’s book [34], or chapter III of Hazewinkel’s book [18], or § W and W Sch . For any N -algebra A , define the N -algebra of Witt vectors with entries in A by W ( A ) := Alg N (Λ N , A ) , and define the N -algebra of Schur–Witt vectors with entries in A by W Sch ( A ) := Alg N (Λ Sch , A ) . The N -algebra structures are inherited from the N - N -bialgebra structure on Λ N and Λ Sch , as explained in section 3. Since both Λ N and Λ Sch are models of Λ Z over06 J. Borger N , both Witt functors agree up to canonical isomorphism with the usual big Wittvector functor on rings.Since Λ N is a composition algebra, W ( A ) has a natural Λ N -action. It is defined,for a ∈ W ( A ) and f ∈ Λ N , by f ( a ) : g a ( g ◦ f ) (6.1.1)for all g ∈ Λ N . Equivalently, the comonad structure map W ( A ) → W ( W ( A )) isgiven by a [ g a ( g ◦ f )]. The analogous statements hold for W Sch and Λ
Sch .Finally, the two functors W and W Sch are related: the inclusion Λ
Sch ⊆ Λ N ofcomposition algebras induces a map W ( A ) −→ W Sch ( A )of Λ Sch -semirings which is natural in A . Consider the diagramΛ
Sch / / / / Λ N N [ h , h , . . . ] O O O O / / / / ❴❴❴ Ψ N O O O O of sub- N -algebras of Λ Z , where the bottom map is dashed to indicate that it isdefined only after base change to Q + . If we apply the functor Alg N ( − , A ) to thisdiagram, we get the following diagram of Witt vectors: W Sch ( A ) σ (cid:15) (cid:15) W ( A ) v o o w (cid:15) (cid:15) tA [[ t ]] A ∞ σ gh o o ❴ ❴ ❴ (6.2.1)where in the bottom row we have made use of the following identifications Alg N ( N [ h , . . . ] , A ) ∼ −→ tA [[ t ]] , a X i a ( h i ) t i Alg N (Ψ N , A ) ∼ −→ A ∞ , a
7→ h a ( ψ ) , a ( ψ ) , . . . i . The dashed arrow indicates that A needs to be a Q + -algebra for σ gh to be defined.Recall from (4.14) that 1 + tA [[ t ]] is in general only a commutative monoid, underusual power-series multiplication; but the other three are N -algebras. (1) v is a morphism of Λ Sch -semirings, and w is a morphismof Ψ N -semirings, and σ and σ gh are morphisms of N -modules: σ ∗ ( x + y ) = σ ∗ ( x ) σ ∗ ( y ) , σ ∗ (0) = 1 . (2) If A is a Z -algebra, then v and σ are bijections. If A is contained in a Z -algebra, then v and σ are injections. itt vectors, semirings, and total positivity If A is a Q -algebra, then w and σ gh are bijections. If A is contained in a Q -algebra, then w and σ gh are injections.Proof. (1): The first statement holds because the inclusions Λ Sch → Λ N and Ψ N → Λ N are morphisms of composition algebras. The second follows from the equations∆ + ( h n ) = P i + j = n h i ⊗ h j and ε + ( h n ) = 0 for n ≥ N , Λ Sch , and N [ h , . . . ] are models for Λ Z over N .(3): Similarly, this holds because Λ N , Ψ N , and N [ h , . . . ] are models for Ψ Q over N . The map v : W ( A ) → W Sch ( A ) can fail to be injective when A isnot contained in a ring. For example, it is not injective when A is the Booleansemiring N / (1 + 1 = 1). See the forthcoming work with Darij Grinberg [7]. When A is a ring, the bijection σ of (6.2.1) allows us to identify Witt vectors and power series. In other words,the complete symmetric functions give a free set of coordinates on the Z -scheme W = Spec (Λ Z ).Note that there are three other common conventions for identifying Witt vectorsand power series. Given any two signs ε , ε ∈ { +1 , − } , the map σ ε ε : W ( A ) ∼ −→ tA [[ t ]] a (cid:0) X i a ( e i )( ε t ) i (cid:1) ε (6.5.1)is an isomorphism of N -modules. We have taken σ = σ −− as our convention. It hasone advantage over the others, which is that σ gh can be defined over Q + ratherthan just Q . Equivalently, the complete symmetric functions can be written aspolynomials in the power sums with nonnegative coefficients. This is not true withthe elementary symmetric functions, even if we allow ourselves to change the signsof half of them. I am not aware of any other convincing reason to prefer one ofthese sign conventions to the others.The presentation Λ Z = Z [ θ , . . . ] gives another set of full coordinates: W ( A ) ∼ −→ A ∞ , a ( a ( θ ) , a ( θ ) , . . . )We will call the elements a ( θ ) , · · · ∈ A the Witt components or Witt coordinates of a . Addition does not have a simple closed form in these coordinates, unlike in theseries coordinates above. But the Witt coordinates do have two advantages over theseries coordinates. First, they are related to the power sums by a simple, closed-form expression (4.5.1). Second, they behave well with respect to localization.More precisely, let E be a set of prime numbers. Then we have a free presentationΛ Z [1 /E ] = Z [1 /E ][ . . . , θ m ◦ ψ n , . . . ] , where m runs over the positive integers whose prime divisors do not lie in E , and n runs over those whose prime divisors all do. (See the isomorphisms (1.20.1) and(5.3.1) in [4], for example.)08 J. Borger
The components of the image of a under the map w : W ( A ) → A ∞ , a
7→ h a ( ψ ) , a ( ψ ) , . . . i are its ghost components . Their advantage is that all algebraic structure is com-pletely transparent—addition and multiplication are performed componentwiseand Frobenius operations are given by scaling the indices. When A is a Q -algebra, w is a bijection, and so the ghost components have no deficiencies over Q . When A is a torsion-free ring (i.e. a flat Z -algebra), w is an injection, and the ghost coor-dinates are still useful there. We write the ghost components with angle bracketsto avoid any confusion with the Witt components. A Witt vector is de-termined by its series coordinates when A is contained in a ring. Indeed, a mapΛ N → A is equivalent to a map a : Λ Z → Z ⊗ N A subject to the effectivity condition a ( m λ ) ∈ A for all partitions λ . Therefore we have W ( A ) = (cid:8) X i a i t i ∈ tA [[ t ]] | X µ M µλ a µ a µ · · · ∈ A (cid:9) , (6.6.1)where M is the transition matrix from the monomial basis ( m λ ) λ to the basis ofmonomials in the complete symmetric functions h n : m λ = X µ M µλ h µ h µ · · · . Formula (6.6.1) also holds for W Sch ( A ), but then M must be the transition matrixfrom the Schur basis to this h basis.Similar statements hold for the ghost components when A is contained in a Q -algebra. For example, we have W Sch ( A ) = (cid:8) h a , a , . . . i ∈ A ∞ | X µ N µλ a µ a µ · · · ∈ A (cid:9) , (6.6.2)where N is the transition matrix from the Schur basis to the basis of monomialsin the power sums: s λ = X µ N µλ ψ µ ψ µ · · · . In this particular case, the matrix N has a well-known description: the block where | λ | = | µ | = n is the inverse of the character table of the symmetric group S n . SeeMacdonald, (7.8) p. 114 [33]. Let us write out theeffectivity conditions on ghost components for W Sch ( A ) and W ( A ) up to weight 4.If a ghost vector h a , . . . , a i ∈ A lies in the image of W Sch ( A ), then the followingeleven elements of Q ⊗ N A are contained in A : a , ( a ± a ) / , ( a ± a a + 2 a ) / , ( a − a ) / , itt vectors, semirings, and total positivity a ± a a + 3 a + 8 a a ± a ) / , (3 a ± a a − a ∓ a ) / , (2 a + 6 a − a a ) / . These are all the conditions corresponding to s λ with | λ | ≤ W ( A ), the following elements are required to be in A : a , a , a , a , ( a − a ) / , a a − a , ( a − a a + 2 a ) / , a a − a , ( a − a ) / , ( a a − a − a a + 2 a ) / , ( a − a a + 3 a + 8 a a − a ) / m λ with | λ | ≤ For general semirings, thereis no purely vector-like description of W ( A ) or W Sch ( A ). Indeed, for any set T , thefunctor A A T is represented by the free N -algebra on T , but by (4.4) and (5.4),neither Λ N nor Λ Sch is free as an N -algebra.Instead Witt vectors are cut out of infinite-dimensional affine space by quadraticrelations determined by the structure constants of multiplication in the relevantbasis. Let P denote the set of partitions. For any λ, µ ∈ P , write m λ m µ = X ν ∈ P b νλµ m ν , s λ s µ = X ν ∈ P c νλµ s ν . Then we have W ( A ) = { a ∈ A P | a = 1 , a λ a µ = X ν b νλµ a ν } W Sch ( A ) = { a ∈ A P | a = 1 , a λ a µ = X ν c νλµ a ν } . Addition and multiplication are then defined using the structure constants for thecoproducts ∆ + and ∆ × with respect to the basis in question. It is often better to view W ( A ) and W Sch ( A )as pro-sets, or pro-discrete topological spaces, as when A is a ring. We do this asfollows. Let ( P i ) i ∈ I denote the filtered system of finitely generated sub- N -algebrasof Λ N . Then we have W ( A ) = Hom(colim i ∈ I P i , A ) = lim i ∈ I Hom( P i , I ) . (6.9.1)So W ( A ), and similarly W Sch ( A ), has the natural structure of a pro-set. When A is a ring, W ( A ) can be expressed as an inverse limit of rings. I do not know,however, whether the analogous statement holds over N : Question 4.
Are W ( A ) and W Sch ( A ) pro-objects in the category of N -algebras?More naturally, is this true as representable functors? Equivalently, can Λ N and Λ Sch be expressed as filtered colimits colim i P i , where each P i is a co- N -algebraobject in Alg N which is finitely generated as an N -algebra? For a stronger form of this question, see (8.1).10
J. Borger
For any N -algebra A ,consider the monoid algebra N [ A ] with the toric Λ N -structure. By the adjunctionproperty of W , the N -algebra map N [ A ] → A defined by [ a ] a lifts to a uniqueΛ N -equivariant map N [ A ] −→ W ( A ) . For a ∈ A , the image of [ a ] in W ( A ) is called the Teichm¨uller lift of a and is alsodenoted [ a ]. Explicitly, [ a ] is the N -algebra map Λ N → A determined by[ a ] : m λ −→ ( a r if λ = ( r )0 otherwise (6.10.1)The Teichm¨uller map a [ a ] is a map of monoids A → W ( A ) under multiplication.For the anti-Teichm¨uller elements, consider the N -algebra N [ η ] / ( η = 1) ⊗ N N [ A ];this has an action of Λ Sch . On the first factor, Λ
Sch acts as in (5.7). On the secondfactor, it acts through the toric action of Λ N . Then the N -algebra map N [ η ] / ( η = 1) ⊗ N N [ A ] −→ A determined by η a ] a for all a ∈ A lifts by adjunction to a uniqueΛ Sch -equivariant map N [ η ] / ( η = 1) ⊗ N N [ A ] −→ W Sch ( A ) . For any a ∈ A , define the anti-Teichm¨uller lift { a } ∈ W Sch ( A ) to be the image of η ⊗ a . So in W ( Z ⊗ N A ) we have { a } = − [ − a ] . (6.10.2)We also have { a } : s λ −→ ( a r if λ = (1 r )0 otherwise (6.10.3)and [ a ] { b } = { ab } , { a }{ b } = [ ab ] , { a } = { } [ a ] . (6.10.4)Observe that the anti-Teichm¨uller lifts exist only in W Sch ( A ) and not generallyin W ( A ). For example, the element { } ∈ W Sch ( N ) is not in the sub- N -algebra W ( N ). Indeed, its ghost vector h , − , , − , . . . i is not even in N ∞ . Let ω : Λ Z −→ Λ Z denote the ring map determined by ω ( h n ) = e n for all n . Then we have ω ( s λ ) = s λ ′ itt vectors, semirings, and total positivity λ , where λ ′ denotes the conjugate partition. (See Macdonald, p. 23 (2.9) ′ [33].)Therefore we have ω (Λ Sch ) = Λ
Sch , and so ω induces a functorial set map W Sch ( A ) −→ W Sch ( A ) , a a ◦ ω. In fact, this map is simply multiplication by { } . Indeed, it follows from theidentities of (4.5.2) that ω ( ψ n ) = ( − n +1 ψ n for all n ; now combine this with the equality { } = h , − , , − , . . . i .The symmetric functions f λ := ω ( m λ ) are sometimes called the forgotten sym-metric functions. Their span Λ for := ω (Λ N ) contains Λ Sch . It represents the functor
Alg N (Λ for , A ) = { } W ( A ) , which is the free W ( A )-module generated by the symbol { } . The induced map { } W ( A ) −→ W Sch ( A )is a W ( A )-module map, but { } W ( A ) cannot be given an N -algebra structuremaking the map an N -algebra map. In particular, Λ for is not a model for Λ Z asco- N -algebra object. N → W Sch ( A ) is injective unless A = 0. Infact, the map ϕ from N to any nonzero Λ Sch -semiring is injective. For if m, n ∈ N have the same image and m < n , then we have0 = ϕ ( (cid:18) mn (cid:19) ) = s n ( ϕ ( m )) = s n ( ϕ ( n )) = ϕ ( (cid:18) nn (cid:19) ) = 1 . So ϕ is injective unless it is the map to the zero ring. In particular, the map from N to any nonzero Λ N -semiring is injective.On the other hand, the Λ Sch -equivariant map N [ η ] / ( η = 1) → W Sch ( A )sending η → { } is not always injective. Indeed, when A = Z / Z , we have[ −
1] = [1] = 1 and hence 1 + η
7. Total positivity
In this section, we will give explicit descriptions of W ( R + ) and W Sch ( R + ) andthen use this to describe W ( N ). These are very rich objects, and there is muchmore to say about them than we can here.We will find it convenient to use the series normalization σ ++ of (6.5), as wellas our standard one σ = σ −− . By (4.5.2), we have σ ( x ) = X i x ( h i ) t i , σ ++ ( x ) = X i x ( e i ) t i . J. Borger
The two are related by the involution f ( t ) f ( − t ) − . In other words, we have σ ++ ( x ) = σ ( { } x ) . (7.0.1) For any Witt vector x ∈ W ( R + ) , write a t + a t + · · · forthe series σ ++ ( x ) ∈ t R + [[ t ]] . Then for all n , we have a n ≤ a n n ! . (7.1.1) In particular, the series P n a n t n converges to an entire function on C .Proof. Since m , n − = e n − e − ne n , we have a n − a − na n ≥ . Then (7.1.1)follows by induction.
A formal series P n a n t n ∈ t R [[ t ]] is said to be totallypositive if all (finite) minors of the infinite matrix ( a i − j ) ij are ≥
0. (To be clear, weunderstand a = 1 and a n = 0 for n <
0. Other authors allow more general series.)For example, up to the 2 × a n ≥ , a n a n + i − j ≥ a n + i a n − j . Chapter 8 of Karlin’s book [21] has an extensive treatment.
A Witt vector x ∈ W ( R ) lies in W Sch ( R + ) if and only if theseries corresponding to x under the bijection σ : W ( R ) → t R [[ t ]] is totallypositive. The same is true for the bijection σ ++ .Proof. First consider the universal series 1 + h t + h t + · · · ∈ Λ Z [[ t ]]. Then it isa standard fact in algebraic combinatorics that the minors of the matrix ( h i − j ) ij generate Λ Sch as an N -module. (The minors are the so-called skew Schur functions s λ/µ , by Macdonald, p. 70 (5.4) [33]. Their N -span contains Λ Sch because everySchur function is a skew Schur function. For the other containment, see (9.1) andthe following text on p. 142 of Macdonald.) Therefore x , viewed as a ring mapΛ Z → R , sends Λ Sch to R + if and only if the corresponding series is totally positive.For the second statement, combine the above with (7.0.1) and the fact that { } is an element of W Sch ( A ) and is invertible. (Edrei [14], Thoma [42]) . The totally positive series in t R [[ t ]] are precisely those of the form e γt Q ∞ i =1 (1 + α i t ) Q ∞ i =1 (1 − β i t ) , (7.4.1) where γ, α i , β i ≥ (and both P i α i and P i β i converge), and every such represen-tation is unique. For a proof, see Karlin’s book [21], in which the result is theorem 5.3, p. 412. itt vectors, semirings, and total positivity
Consider a Witt vector x ∈ W ( R ) . Then x lies in W Sch ( R + ) ifand only if σ ++ ( x ) is of the form e γt Q ∞ i =1 (1 + α i t ) Q ∞ i =1 (1 − β i t ) , (7.5.1) where γ, α i , β i ≥ . Similarly, x lies in W ( R + ) if and only if σ ++ ( x ) is of the form e γt ∞ Y i =1 (1 + α i t ) , (7.5.2) where γ, α i ≥ .Proof. The first part follows from (7.3) and (7.4).Now consider the second part. The first part and (7.1) imply that for every Wittvector x ∈ W ( R + ), the series σ ++ ( x ) is of the form (7.5.2). Conversely, because wehave 1 + αt = σ ++ ([ α ]) and e γt = lim n →∞ (cid:16) γtn (cid:17) n , any series of the form (7.5.2) is a limit of finite products of series in σ ++ ( W ( R + )).Now observe that σ ++ identifies W ( R + ) with a submonoid of 1 + t R + [[ t ]]. Furtherit is a closed subset because it is defined by a family of nonnegativity conditions,as in (6.6.1). Therefore any series of the form (7.5.2) lies in the image of σ ++ .(Compare Kingman [23, 24].) W ( R + ) and W Sch ( R + ) as convergent monoid algebras. Equivalently, the subset W Sch ( R + ) consists of the Witt vectors in W ( R ) that canbe represented (necessarily uniquely) in the form ∞ X i =1 [ α i ] + ∞ X i =1 { β i } + [ γ ] ξ, where γ, α i , β i ≥ P i α i and P i β i converge) and where ξ is the Wittvector with ghost components h , , , . . . i , or equivalently such that σ ( ξ ) = e t .Another interpretation is that the evident map is an isomorphism N [ R + ] b ⊗ N ( N [ η ] / ( η = 1)) ⊕ R + ξ ∼ −→ W Sch ( R + ) . Similarly, we have W ( R + ) = N [ R + ] b ⊕ R + ξ . W ( R + ) and entire functions. If we view (7.5.2) as a Hadamardfactorization, then we see that yet another interpretation of W ( R + ) is that it isthe set of entire functions f on C of order at most 1 such that f (0) = 1, the zerosof f are negative real numbers, and p = 0 in the notation of lecture 4 of Levin’sbook [30].14 J. Borger W ( N ) , viewed as a subset of t Z [[ t ]] by the map σ ++ , agrees withthe set of polynomials in t Z [ t ] whose complex roots are all real and negative.In particular, W ( N ) is countable.Proof. We have W ( N ) = W ( Z ∩ R + ) = W ( Z ) ∩ W ( R + ) . By (7.5), elements onthe right-hand side correspond to series of the form (7.5.2) with coefficients in Z .Certainly this includes all the polynomials in 1+ t Z [ t ] with only negative real roots.Conversely, the coefficients of such a series tend to 0, by (7.1); so all such seriesare polynomials. We can reinterpret this in a way that treats the finite and infiniteplaces of Q as similarly as possible. A monic p -adic polynomial has coefficientsin Z p if and only if all its roots are integral over Z p . Therefore W ( N ), viewedas a subset of 1 + t Q [[ t ]] via σ ++ , is the set of polynomials that when written as Q i (1 + α i t ), have the property that every α i is integral at each finite place and isreal and positive at the infinite place.Rephrasing again, if O tp¯ Q denotes the multiplicative monoid of algebraic numberswhich are integral at all finite places and which are real and positive at all infiniteplaces, then we have W ( N ) = N [ O tp¯ Q ] Gal(¯ Q / Q ) . (7.9.1) W does not preserve surjectivity. Indeed, N sur-jects onto a nonzero ring, for instance Z / Z . But W ( N ) is countable while W applied to any nonzero ring is uncountable. Of course, W does preserve surjectiv-ity for maps between rings, by (6.5).In fact, W Sch ( N ) is also countable. This will be shown in forthcoming workwith Darij Grinberg [7]. It follows that W Sch does not preserve surjectivity either.
There is a multi-dimensional generalization of (7.8). Let A be adiscrete subring of R n , for some n ≥
0, and write A + = A ∩ R n + . Then W ( A + ),viewed as a subset of W ( A ) = 1 + tA [[ t ]], consists of the polynomials which splitcompletely over R n + . In particular, this applies to any totally real number field K ,in which case we have the following generalization of (7.9.1): W ( O K, + ) = N [ O tp¯ Q ] Gal(¯ Q /K ) . (7.11.1)In particular, there are countably many Witt vectors with entries in the algebraof algebraic integers that are real and nonnegative at all infinite places.
8. A model for the p -typical symmetric functions over N p -typical Witt vectors and general truncation sets. Following Bergman(lecture 26 of [34]), let us say a set S of positive integers is a truncation set if it isclosed under taking divisors. For any truncation set S , writeΛ Z ,S := Z [ θ d | d ∈ S ] ⊆ Z [ θ , θ , . . . ] = Λ Z , itt vectors, semirings, and total positivity θ d are the Witt symmetric functions of (4.5). For any ring A , write W S ( A ) := Alg Z (Λ Z ,S , A )for the corresponding ring of Witt vectors.The induced map W ( A ) → W S ( A ) is surjective for all A . Indeed, any retractionΛ Z → Λ Z ,S gives a functorial section. The quotient W S ( A ) of W ( A ) is in fact aquotient ring, or equivalently Λ Z ,S is a sub- Z - Z -algebra of Λ Z . One can show thisas follows: By induction on S , the ring Q ⊗ Z Λ Z ,S agrees with Q [ ψ d | d ∈ S ], whichis a sub- Q - Z -algebra of Ψ Q ; therefore ∆ + (Λ Z ,S ) is contained in Λ ⊗ Z ∩ ( Q ⊗ Z Λ Z ,S ) ⊗ and hence Λ ⊗ Z ,S .This functor W S : Alg Z → Alg Z then agrees with the usual one in Bergman [34],at least up to canonical isomorphism. Similarly, it agrees with the Witt vectorfunctors of [4], as long as S is of the form { d | d divides m } for some integer m ≥
1. (The functors in [4] are defined only in that context. Note however thatsuch truncation sets form a cofinal family.) More precisely, W S and Λ Z ,S agree withthe objects W Z ,E,n and Λ Z ,E,n defined in [4], section 1, where E denotes the set ofprime divisors of m , and n ∈ N E is the vector with components n p = ord p ( m ).When S = { , p, . . . , p k } for some prime p , the functor W S is the p -typical Wittvector functor of length k (or more traditionally k +1) discussed in the introduction.In this case, we will write W ( p ) ,k ( A ) and Λ Z , ( p ) ,k instead of W S and Λ Z ,S . When k = ∞ , we will also write W ( p ) and Λ Z , ( p ) . Question 5.
Does Λ Z ,S have a model Λ N ,S over N as a co- N -algebra object? If so,are there models such that Λ N ,S ◦ Λ N ,S ′ ⊆ Λ N ,SS ′ , where SS ′ = { ss ′ | s ∈ S, s ′ ∈ S ′ } ? The purpose of this section is to show the answers are yes in the p -typical case,when S and S ′ contain only powers of a single prime p . p -typical symmetric functions. Let p be a prime number. Re-call the notation d p := e p − ψ p p = m p − , + · · · + ( p − m p ∈ Λ N ∩ Λ Z , ( p ) , . Let Λ N , ( p ) ,k denote the sub- N -algebra of Λ Z , ( p ) ,k generated by the set { ψ ◦ ip ◦ d ◦ jp | i + j ≤ k } , and write A k = N [ x i,j | ≤ i + j ≤ k ] / ( x pi,j = x i +1 ,j + px i,j +1 | i + j ≤ k − . So A k is an algebra over N [ x i,j | i + j = k ]; as a module, it is free of rank p k ( k +1)2 . The N -algebra map N [ x i,j | i + j ≤ k ] → Λ N , ( p ) ,k sending x i,j to ψ ◦ ip ◦ d ◦ jp factors through A k , and the induced map is an isomorphism A k ∼ −→ Λ N , ( p ) ,k . (8.3.1) In particular, N [ ψ ◦ ip ◦ d ◦ jp | i + j = k ] is freely generated as an N -algebra by the k + 1 elements ψ ◦ ip ◦ d ◦ jp , and Λ N , ( p ) ,k is a free module of rank p k ( k +1)2 over it. J. Borger
Proof.
A morphism ϕ : A k → Λ N , ( p ) ,k sending x i,j to ψ ◦ ip ◦ d ◦ jp exists because wehave ( ψ ◦ ip ◦ d ◦ jp ) p = e p ◦ ( ψ ◦ ip ◦ d ◦ jp )= ( ψ p + pd p ) ◦ ( ψ ◦ ip ◦ d ◦ jp )= ψ ◦ ( i +1) p ◦ d ◦ jp + pψ ◦ ip ◦ d ◦ ( j +1) p . (8.3.2)since ψ p commutes under composition with every element. Also, ϕ is clearly sur-jective.Let us now show injectivity. Since A k is a free N -module, it is enough to do soafter tensoring with Q . Consider the diagram Q [ x , , . . . , x k, ] / / incl / / Q ⊗ N A k id ⊗ ϕ / / Q ⊗ N Λ N , ( p ) ,k . Observe that the composition is injective because it sends the x i, to the ψ ◦ ip ,which are algebraically independent elements of Q ⊗ Z Λ Z , ( p ) ,k (and in fact are free Q -algebra generators). Therefore it is enough to show that the first inclusion isan equality. That is, it is enough to show that the elements x , , . . . , x k, generate Q ⊗ N A k as a Q -algebra.This is follows directly from the relations x pi,j = x i +1 ,j + px i,j +1 . Indeed theyimply that, for any j , if all x i,j lie in a given sub- Q -algebra, then so do all x i,j +1 .Therefore, the sub- Q -algebra generated by all x i, contains all x i, , and hence all x i, , and so on. Therefore it consists of all of Q ⊗ N A . (1) Λ N , ( p ) ,k is free as an N -module. (2) It is an N -model for Λ Z , ( p ) ,k as a co- N -algebra object in a unique way. (3) We have Λ N , ( p ) ,k ◦ Λ N , ( p ) ,k ′ ⊆ Λ N , ( p ) ,k + k ′ .Proof. (1): This follows from (8.3).(2): The induced map Z ⊗ N Λ N , ( p ) ,k → Λ Z , ( p ) ,k is an injection, and it is asurjection because Λ Z , ( p ) ,k is generated as a Z -algebra by e, d p , . . . , d ◦ kp , all of whichare contained in Λ N , ( p ) ,k . Therefore Λ N , ( p ) ,k is a model for Λ Z , ( p ) ,k as an algebra.To show it is a model as a co- N -algebra object, it is enough to show∆( x ) ∈ Λ N , ( p ) ,k ⊗ N Λ N , ( p ) ,k (8.4.1)as x runs over a set of N -algebra generators for Λ N , ( p ) ,k , and as ∆ runs over thetwo coproducts ∆ + , ∆ × .First consider the case k = 1, where we have the generators e, ψ p , d p . Since e and ψ p are both N -algebra-like elements, (8.4.1) holds for them. For d p , it followsfrom the positivity of the coefficients in the following equalities:∆ + ( d p ) = d p ⊗ ⊗ d p + e ⊗ e ∆ × ( d p ) = d p ⊗ ψ p + ψ p ⊗ d p + pd p ⊗ d p . itt vectors, semirings, and total positivity N , ( p ) , is a co- N -algebra object in Alg N ,we can form (Λ N , ( p ) , ) ⊙ k . Consider the map(Λ N , ( p ) , ) ⊙ k −→ (Λ Z , ( p ) , ) ⊙ k −→ Λ Z , ( p ) ,k of co- N -algebra objects defined by f ⊙ · · · ⊙ f k f ◦ · · · ◦ f k . Then its image isequal to Λ N , ( p ) ,k . Indeed, Λ N , ( p ) , is generated by e, ψ p , d p , all of which commutewith each other under composition; so the image is the sub- N -algebra generatedby all ◦ -words in e, ψ p , d p of length k , which is Λ N , ( p ) ,k . Thus we have a surjection(Λ N , ( p ) , ) ⊙ k −→ Λ N , ( p ) ,k (8.4.2)of co- N -algebra objects, and hence a diagram(Λ N , ( p ) , ) ⊙ k / / ∆ (cid:15) (cid:15) Λ Z , ( p ) ,k ∆ (cid:15) (cid:15) (Λ N , ( p ) , ) ⊙ k ⊗ N (Λ N , ( p ) , ) ⊙ k / / Λ Z , ( p ) ,k ⊗ N Λ Z , ( p ) ,k Therefore ∆(Λ N , ( p ) ,k ) is contained in the image of (Λ N , ( p ) , ) ⊙ k ⊗ N (Λ N , ( p ) , ) ⊙ k ,which is contained in Λ N , ( p ) ,k ⊗ N Λ N , ( p ) ,k . Therefore Λ N , ( p ) ,k is a model as a co- N -algebra object.(3): Because (8.4.2) is surjective, Λ N , ( p ) ,k ◦ Λ N , ( p ) ,k ′ equals the image of(Λ N , ( p ) , ) ⊙ k ⊙ (Λ N , ( p ) , ) ⊙ k ′ −→ Λ Z , ( p ) which is Λ N , ( p ) ,k + k ′ . p -typical Witt vectors and Λ-structures for semirings. We can puteverything above together in what should now be a familiar way.Define Λ N , ( p ) := colim k Λ N , ( p ) ,k . It is a flat model for Λ Z , ( p ) over N , as composition algebra. Write W ( p ) ( A ) := Alg N (Λ N , ( p ) , A )for its Witt vector functor. The truncated version is W ( p ) ,k ( A ) := Alg N (Λ N , ( p ) ,k , A ) , and we have W ( p ) ( A ) = lim k W ( p ) ,k ( A ) . These Witt functors take values in N -algebras. If A is a ring, then they agree withthe usual p -typical Witt vector rings.18 J. Borger
Giving an action of Λ N , ( p ) on an N -algebra A is equivalent to giving an N -algebraendomorphism ψ p : A → A and a set map d p : A → A satisfying the identities ψ p ( d p ( x )) = d p ( ψ p ( x )) ψ p ( x ) + pd p ( x ) = x p d p ( x + y ) = d p ( x ) + d p ( y ) + p − X i =1 p (cid:18) pi (cid:19) x i y p − i d p ( xy ) = d p ( x ) ψ p ( y ) + ψ p ( x ) d p ( y ) + pd p ( x ) d p ( y ) d p (0) = 0 d p (1) = 0 . (8.5.1)Observe that if A is flat over N , then ψ p determines d p , assuming it exists.This is because additive cancellativity and p -cancellativity are flat-local proper-ties, by (2.9). Thus giving a Λ N , ( p ) -structure on a flat N -algebra is equivalent togiving an N -algebra endomorphism ψ p lifting the Frobenius map on Z /p Z ⊗ N A and satisfying ψ p ( x ) ≤ x p for all x . Here we write a ≤ b if there exists an element c ∈ A such that a + c = b . It is possibleto give an interpretation of the bases of Λ N , ( p ) ,k and Λ N , ( p ) in the language ofpartitions. For simplicity of notation, let us write Λ N , ( p ) , ∞ = Λ N , ( p ) and henceallow k = ∞ .Over Z , there is a basis of Λ Z , ( p ) ,k given by monomials in the d ◦ ip or the θ p i ,where i ≥
0. They can be indexed by usual partitions λ that are p -typical in thesense that all parts λ j are powers of p . For example, one could use the family Q j θ λ j . Equivalently, we could index them by the p -typical multiplicity vectors m ∈ L i ≥ N , where m i is the number of j such that λ j = p i .On the other hand, the N -basis of Λ N , ( p ) ,k consists of monomials Y i + j ≤ k ( ψ ◦ ip ◦ d ◦ jp ) m ij , such that when i + j < k , we have m ij < p . Observe that such a monomial remainsa basis element in Λ N , ( p ) ,k +1 if and only if m ij < p when i + j = k .We might think of the vector m ∈ L i,j ≥ N as the multiplicity vector of a2-dimensional p -typical partition. Such a partition would be an expression ofthe form P i,j m ij p i + j . The basis for Λ N , ( p ) ,k would then be indexed by all 2-dimensional p -typical partitions subject to the conditions that there are less than p parts ( i, j ) when i + j < k , and no parts ( i, j ) when i + j > k . Since ψ p , d p ∈ Λ N , we haveΛ N , ( p ) ⊆ Λ N . This induces canonical algebra maps W ( A ) → W ( p ) ( A ) for all N -algebras A . In particular, for each a ∈ A , there is a Teichm¨uller lift [ a ] ∈ W ( p ) ( A ).It is the image of the usual Teichm¨uller lift [ a ] ∈ W ( A ). itt vectors, semirings, and total positivity N , ( p ) Λ Sch , simply because ψ p Λ Sch . Inparticular, there is no functorial map W Sch ( A ) → W ( p ) ( A ) that agrees with theusual one for rings. W ( p ) ,k ( A ). The presentation (8.3.1) trans-lates directly into finite descriptions of the Witt vectors of finite length: W ( p ) ,k ( A ) = (cid:8) ( a i,j ) ∈ A { ( i,j ) | i + j ≤ k } | a pi,j = a i +1 ,j + pa i,j +1 for i + j < k (cid:9) . For example, W ( p ) , ( A ) = (cid:8) ( a , a , a ) | a p = a + pa (cid:9) In general, W ( p ) ,k = Spec (Λ N , ( p ) ,k ) is the locus in the N -scheme A ( k +22 ) N defined bythe (cid:0) k +12 (cid:1) relations in the algebra A k of (8.2).As usual with Witt vectors, the algebraic structure is not transparent whenexpressed in coordinates. The simplest nontrivial example is W ( p ) , ( A ), where wehave( a , a , a ) + ( b , b , b ) = ( a + b , a + b , a + b + p − X i =1 p (cid:18) pi (cid:19) a i b p − i )( a , a , a )( b , b , b ) = ( a b , a b , a b + a b + pa b )0 = (0 , , , , . This is just another expression of the formulas of (8.5.1). W ( p ) ,k ( A ) when A is contained in a ring. In this case, we can ignorethe relations in the presentation (8.3.1) and instead describe W ( p ) ,k ( A ) in termsof the usual p -typical Witt vector ring W ( p ) ,k ( Z ⊗ N A ) and effectivity conditionscorresponding to the generators ψ ◦ ip ◦ d ◦ jp . Indeed, a morphism Λ N , ( p ) ,k → A isequivalent to a morphism a : Λ Z , ( p ) ,k → Z ⊗ N A such that a ( ψ ◦ ip ◦ d ◦ jp ) ∈ A for all i, j . Thus we have W ( p ) ,k ( A ) = (cid:8) a ∈ W ( p ) ,k ( Z ⊗ N A ) | a ( ψ ◦ ip ◦ d ◦ jp ) ∈ A for i + j ≤ k (cid:9) . For instance, if A is contained in a Q -algebra, then this permits a recursive de-scription in terms of ghost components: W ( p ) ,k ( A ) is the set of ghost vectors h a , . . . , a k i ∈ A k +1 satisfying the following property: h a , . . . , a k i , D a p − a p , . . . , a pk − − a k p E ∈ W ( p ) ,k − ( A ) . J. Borger
Thus the conditions are that for all i ≥
0, the elements a pi − a i +1 p , p (cid:16) a pi − a i +1 p (cid:17) p − p (cid:16) a pi +1 − a i +2 p (cid:17) , p p (cid:16) a pi − a i +1 p (cid:17) p − p (cid:16) a pi +1 − a i +2 p (cid:17)! p − p p (cid:16) a pi +1 − a i +2 p (cid:17) p − p (cid:16) a pi +2 − a i +3 p (cid:17)! and so on lie in A .For example, W ( p ) , ( R + ) is the set of ghost vectors h x, y i in R with y ≤ x p .Similarly one can show that W ( p ) , ( R + ) is the set of ghost vectors in R of theform h a, a p x, a p y i , where x ≤ ≤ x p − y ≤ (1 − x ) p p p − . Thus the pre-image of 1 under the projection W ( p ) , ( R + ) → R + onto the firstcoordinate is the 2-simplex bounded by the curves y = x p , y = x p − (1 − x ) p /p p − and y = 0 in the xy -plane. W ( p ) ,k +1 ( A ) → W ( p ) ,k ( A ) isnot generally surjective. It is surjective when A is a ring or when k = 0. Butwhen k ≥ A is general, it is not. It is enough to check this in the universalcase, when A = Λ N , ( p ) ,k . In other words, it is enough to show the inclusionΛ N , ( p ) ,k → Λ N , ( p ) ,k +1 has no retraction in the category Alg N . So suppose ϕ is sucha retraction. By (8.3.2), we have( ψ ◦ ip ◦ d ◦ jp ) p = ψ ◦ ( i +1) p ◦ d ◦ jp + pψ ◦ ip ◦ d ◦ ( j +1) p . Now suppose i + j = k . Then since ϕ is a retraction, we have( ψ ◦ ip ◦ d ◦ jp ) p = ϕ ( ψ ◦ ip ◦ d ◦ jp ) p = ϕ ( ψ ◦ ( i +1) p ◦ d ◦ jp ) + pϕ ( ψ ◦ ip ◦ d ◦ ( j +1) p ) . But by (8.3), the left-hand side is additively indecomposable in Λ N , ( p ) ,k , and soboth terms on the right-hand side vanish. So we have ϕ ( ψ ◦ ip ◦ d ◦ ( j +1) p ) = 0 whenever i + j = k . Taking i = k − j = 1 gives ϕ ( ψ ◦ ( k − p ◦ d p ) p = ϕ ( ψ ◦ kp ◦ d p ) + pϕ ( ψ ◦ ( k − p ◦ d ◦ p ) = 0 . But this is impossible because ϕ is a retraction. In the theory of Λ-rings, a finiteprime p allows us to speak of two things: p -adic integrality, which is a property,and Frobenius lifts at p , which are structures. The fundamental point of thischapter is that it is reasonable for some purposes to think of positivity as p -adicintegrality at the place p = ∞ . So the infinite prime plays the first role here butnot the second. This is meager when compared to the rich role the infinite prime itt vectors, semirings, and total positivity p = ∞ and whether there is an ∞ -typical theory that can be isolated from therest of the primes. One might hope that flows will appear here.
9. On the possibility of other models
So far, we have not been concerned with whether our N -models are the mostnatural ones—their existence has been interesting enough. The purpose of thisshort section is to raise some questions in this direction.As discussed in (5.11), the connection between symmetric functions and arith-metic algebraic geometry is explained by Wilkerson’s theorem, which we interpretas saying that Λ Z is the composition ring that controls commuting Frobenius lifts.It is natural to ask whether there are similar, arithmetically satisfying descriptionsof the composition algebras over N we have considered. As explained in (8.5),there is a such a description in the p -typical case. It would be interesting to findone for Λ N or Λ Sch . A less satisfying alternative would be to single out Λ N andΛ Sch among all N -models by some general properties, and at least this form of thequestion admits a precise expression: Question 6.
Are Λ N , Λ Sch , and possibly the B p (of question 3 in 5.11) the onlyflat models for Λ Z over N ? If not, is Λ Sch the minimal one? Is Λ N the maximalone? Is Λ N , ( p ) the only flat model for Λ Z , ( p ) over N ? Whether we ask for models as composition algebras or models as co- N -algebraobjects, I do not know the answer. I do not even know the answer to analogousquestions about integrality at the finite primes. For instance, is Λ Z , ( p ) the maximalintegral model for Z [1 /p ] ⊗ Z Λ Z , ( p ) ? Question 7.
Over Q + , there is another model for Λ Q , namely Ψ Q + . Is there stillanother? Question 8.
Let K be a number field, and let T be a set of embeddings K → R .Do the composition algebras Λ O K ,E of (5.10) have models over the sub- N -algebraof O K consisting of elements that are nonnegative under all σ ∈ T ? We have seen that if K = Q and T consists of the unique embedding, theanswer is yes in two cases: when E consists of all maximal ideals of Z or when itconsists of only one. I do not know the answer in any other case, unless T or E isempty. k -Schur functions and truncated Witt vectors Let Λ Z ,k denote Z [ h , . . . , h k ]. Thus, in the notation of (8.1), we have Λ Z ,k = Λ Z ,S ,where S is the truncation set { , , . . . , k } . The purpose of this section is to show22 J. Borger how k -Schur functions, a recent development in the theory of symmetric functions,allow us to give an N -model Λ Sch k for Λ Z ,k which approaches Λ Sch as k tends toinfinity. Unfortunately, Λ Sch k is only a model as a co- N -module object, and not asa co- N -algebra object. This would seem to be fatal for any application of k -Schurfunctions to Witt vectors as objects of arithmetic algebraic geometry. But they dohave several properties that are good from the point of view of Witt vectors, andthere are several parallels with the p -typical N -models of section 8. The purposeof this humble section is just to enter the details into the literature, in case theycan be of use to anyone else. k -Schur functions and Λ Sch k . It is not possible to make a Z -basis forΛ Z ,k out of usual Schur functions. This is because there are only finitely manySchur functions in any given Λ Z ,k . But Lapointe–Lascoux–Morse [28] discoveredcertain symmetric functions that form a basis for Λ Z ,k and are similar to Schurfunctions in many ways. They call them k -Schur functions and denote them s ( k ) λ ,where λ runs over all partitions ( λ , . . . ) which are k -bounded in the sense that λ ≤ k .Our reference for k -Schur functions will be the book [25] by Lam et al. , inparticular part 2, which was written by Morse, Schilling, and Zabrocki and isbased on lectures by Lapointe and Morse. They consider more than one definitionof k -Schur function but conjecture that they all agree. For definiteness, we willtake s ( k ) λ := s ( k ) λ [ X ; 1]as our definition, where s ( k ) λ [ X ; t ] is defined there in equation (3.16), p. 81. See pp.83–84 for a discussion of the relations with the other definitions.Define Λ Sch k := M λ N s ( k ) λ . As explained in part 2, section 4.5 of [25], the family s ( k ) λ forms a Z -basis for Λ Sch k .In other words, Λ Sch k is free over N and is a model for Λ Z ,k over N as a module. For k ≥ , we have (1) Λ Sch k is a model for Λ Z ,k as a co- N -module object in Alg N , (2) s ( k ) λ = s λ , if k ≥ λ + l − , where l is the length of λ , (3) Λ Sch k ⊆ Λ Sch k +1 and Λ Sch k ⊆ Λ Sch , (4) Λ Sch k is finitely presented as an N -algebra.Proof. These are mostly just restatements of results collected in part 2, chapter 4of the book [25].(1): This follows from corollaries 8.1 and 8.2 of Lam [27]. See sections 4.7 and4.8 of part 2 of [25]. itt vectors, semirings, and total positivity k ≥ P i λ i is discussed in part 2, section 4.1 of [25].(3): The first statement is theorem 2 of Lam–Lapointe–Morse–Shimozono [26].See section 4.10 of part 2 of [25]. The second statement follows from the first,together with part (2).(4): This follows from the multiplication rule established in theorem 40 ofLapointe–Morse [29]. See section 4.6 of part 2 of [25]. As a module over the N -algebra generated by the k k -rectangular k -Schur functions, Λ Sch k is freely generatedby the k ! k -irreducible k -Schur functions. Further, the k -rectangular k -Schur func-tions are algebraically independent. So Λ Sch k can be generated by k ! + k elementswith (cid:0) k !+12 (cid:1) quadratic relations. There are some similarities between Λ
Sch k and Λ N , ( p ) ,k . Comparethe preceding proposition with section 8, and especially the presentation of Λ Sch k mentioned in the proof of (4) above with the presentation of (8.3.1). For any N -algebra A ,define W Sch k ( A ) := Alg N (Λ Sch k , A ) . It follows from (10.2)(4) that W Sch k ( A ) can be described as the subset of a finite-dimensional affine space A N satisfying a finite list of equations.By (10.2)(2)–(3), we have Λ Sch = colim k Λ Sch k and W Sch ( A ) = lim k W Sch k ( A ).It follows from (10.2)(1) that W Sch k ( A ) inherits an N -module structure and that W Sch k ( A ) = W k ( A ) when A is a ring. Unlike in the case when A is a ring, W Sch k ( A )does not generally inherit an N -algebra structure. Sch k is not a co- N -algebra object. It is for k ≤ × ( s (3)22 ) Λ Sch3 ⊗ Λ Sch3 , and so it failsfor k = 3. This can be checked by hand using the following equalities:12 s (3)22 = ψ + 3 ψ − ψ ψ ψ = s (3)1111 + 2 s (3)211 + 2 s (3)22 + s (3)31 ψ = s (3)1111 − s (3)211 + 2 s (3)22 + s (3)31 ψ ψ = s (3)1111 − s (3)211 − s (3)22 + s (3)31 . On the other hand, ∆ × ( s (3)22 ) is contained in Λ Sch4 ⊗ Λ Sch4 . This is just because itis an N -linear combination of elements of the form s λ ⊗ s µ , where λ and µ arepartitions of 2 + 2 = 4; so we have s λ = s (4) λ and s µ = s (4) µ for all λ, µ in question.It is also not true that Λ Sch k ◦ Λ Sch l ⊆ Λ Sch kl for all k, l . According to my com-putations, it is true if k, l ≤ k, l ) = (3 , k = l = 3, it fails: thecoefficient of s (9)6331111 in s (3)22 ◦ s (3)22 is −
1. In fact, we have Λ
Sch3 ◦ Λ Sch3 Λ Sch11 andΛ
Sch3 ◦ Λ Sch3 ⊆ Λ Sch12 .24
J. Borger
11. Remarks on absolute algebraic geometry
This volume is a collection of contributions on the theme of the mythical fieldwith one element. One can see this chapter from that point of view, although Ihave so far avoided making the connection. There are two natural approaches torigidifying the category of rings—one can look for models over N , or one can addstructure, such as a Λ Z -ring structure, which we think of as descent data to theabsolute point [3]. In this chapter, we have combined the two. I do not have muchmore to say about the philosophy of the field with one element than I already havesaid in [3], but this way of thinking does suggest some mathematical questions. Question 9.
Is it possible to extend the constructions W and W Sch to non-affine N -schemes? What about their adjoints A Λ N ⊙ A and A Λ Sch ⊙ A ? Over Z , this was done in my paper [5], but there are several complications over N . The most important is that over Z I used Witt vectors of finite length, becauseit is better to think of W ( A ) as a projective system of discrete rings, rather thanactually taking the limit. But there is not yet any finite-length version of the bigWitt vector functor for N -algebras. On the other hand, we do have finite-length p -typical functors for N -algebras available; so it is probably easier to make immediateprogress there.A similar question is whether the concept of a Λ N -structure or Λ Sch -structurecan be extended to nonaffine N -schemes. Over Z , this is done using the functor W ∗ = colim n W ∗ n , where W ∗ n is the extension of W n to nonaffine schemes. (Onecould also use its right adjoint W ∗ = lim n W n ∗ .) So the two questions are indeedclosely related. The following question is a natural guide: Question 10.
Let X be a Λ N -scheme that is flat and locally of finite presentationover N . Does there exist a toric variety Y over N and a surjective Λ N -morphism Y → X ? This requires some explanation. By a toric variety over N , I mean an N -schemethat can be formed by gluing together affine N -schemes of the form Spec ( N [ M ]),where M is a commutative monoid, along other schemes of the same form, where allthe gluing maps are induced by maps of the monoids. Surjectivity of a morphismof N -schemes can be understood in the sense of the Zariski topos. Finally, while aΛ N -structure has at the moment no precise meaning for nonaffine X , it is possibleto strengthen the question to a precise one that is still open. For instance, we canrequire only that the base change of X to Z admit a Λ Z -structure. itt vectors, semirings, and total positivity References [1] G. M. Bergman and A. O. Hausknecht,
Co-groups and co-rings in categories ofassociative rings , volume 45 of
Mathematical Surveys and Monographs . AmericanMathematical Society, Providence, RI, 1996.[2] F. Borceux,
Handbook of categorical algebra. 2 , volume 51 of
Encyclopedia of Math-ematics and its Applications . Cambridge University Press, Cambridge, 1994. Cate-gories and structures.[3] J. Borger, Λ-rings and the field with one element. arXiv:0906.3146v1.[4] J. Borger, The basic geometry of Witt vectors, I: The affine case.
Algebra NumberTheory (2011), 231–285.[5] J. Borger, The basic geometry of Witt vectors. II: Spaces. Math. Ann. (2011),877–933.[6] J. Borger and B. de Smit, Galois theory and integral models of Λ-rings.
Bull. Lond.Math. Soc. (2008), 439–446.[7] J. Borger and D. Grinberg. Boolean Witt vectors and an integral Edrei-Thomatheorem. arXiv:1311.5031v2 [8] J. Borger and B. Wieland, Plethystic algebra. Adv. Math. (2005), 246–283.[9] N. Bourbaki, ´El´ements de math´ematique. Alg`ebre commutative. Chapitres 8 et 9 .Springer, Berlin, 2006. Reprint of the 1983 original.[10] A. Buium, Arithmetic analogues of derivations.
J. Algebra (1997), 290–299.[11] A. Buium,
Arithmetic differential equations , volume 118 of
Mathematical Surveysand Monographs . American Mathematical Society, Providence, RI, 2005.[12] A. Connes and C. Consani, Characteristic 1, entropy and the absolute point
Non-commutative geometry, arithmetic, and related topics.
Johns Hopkins Univ. Press,Baltimore, MD (2011), 75–139[13] W. F. Doran, IV, A proof of Reutenauer’s − q ( n ) conjecture. J. Combin. Theory Ser.A (1996), 342–344.[14] A. Edrei, On the generating functions of totally positive sequences. II. J. AnalyseMath. (1952), 104–109.[15] V. Fock and A. Goncharov, Moduli spaces of local systems and higher Teichm¨ullertheory. Publ. Math. Inst. Hautes ´Etudes Sci. (2006), 1–211.[16] J.-M. Fontaine, Le corps des p´eriodes p -adiques. Ast´erisque (1994), 59–111.With an appendix by Pierre Colmez, P´eriodes p -adiques (Bures-sur-Yvette, 1988).[17] J. S. Golan, The theory of semirings with applications in mathematics and theoreticalcomputer science , volume 54 of
Pitman Monographs and Surveys in Pure and AppliedMathematics . Longman Scientific & Technical, Harlow, 1992.[18] M. Hazewinkel,
Formal groups and applications , volume 78 of
Pure and AppliedMathematics . Academic Press Inc. [Harcourt Brace Jovanovich Publishers], NewYork, 1978.[19] L. Hesselholt and I. Madsen, On the K -theory of local fields. Ann. of Math. (2) (2003), 1–113.[20] L. Illusie, Complexe de de Rham-Witt et cohomologie cristalline.
Ann. Sci. ´EcoleNorm. Sup. (4) (1979), 501–661. J. Borger[21] S. Karlin,
Total positivity. Vol. I . Stanford University Press, Stanford, Calif, 1968.[22] Y. Katsov, On flat semimodules over semirings.
Algebra Universalis (2004),287-299.[23] J. F. C. Kingman, The representation of partition structures. J. London Math. Soc.(2) (1978), 374–380.[24] J. F. C. Kingman, The coalescent. Stochastic Process. Appl. (1982), 235–248.[25] T. Lam, L. Lapointe, J. Morse, A. Schilling, M. Shimozono, and M. Zabrocki, k -schurfunctions and affine schubert calculus. arXiv:1301.3569v1.[26] T. Lam, L. Lapointe, J. Morse, and M. Shimozono, k-shape poset and branching ofk-schur functions. Mem. Amer. Math. Soc.
To appear.[27] T. Lam, Schubert polynomials for the affine Grassmannian.
J. Amer. Math. Soc. (2008), 259–281.[28] L. Lapointe, A. Lascoux, and J. Morse, Tableau atoms and a new Macdonald posi-tivity conjecture. Duke Math. J. (2003), 103–146.[29] L. Lapointe and J. Morse, A k -tableau characterization of k -Schur functions. Adv.Math. (2007), 183–204.[30] B. Ya. Levin,
Lectures on entire functions , volume 150 of
Translations of Mathe-matical Monographs . American Mathematical Society, Providence, RI, 1996. In col-laboration with and with a preface by Yu. Lyubarskii, M. Sodin and V. Tkachenko,Translated from the Russian manuscript by Tkachenko.[31] O. Lorscheid, The geometry of blueprints: Part I: Algebraic background and schemetheory.
Adv. Math. (2012), 1804–1846.[32] G. Lusztig, A survey of total positivity.
Milan J. Math. (2008), 125–134.[33] I. G. Macdonald, Symmetric functions and Hall polynomials . Oxford Mathemati-cal Monographs. The Clarendon Press Oxford University Press, New York, secondedition, 1995. With contributions by A. Zelevinsky, Oxford Science Publications.[34] D. Mumford,
Lectures on curves on an algebraic surface . With a section by G.M. Bergman. Annals of Mathematics Studies, No. 59. Princeton University Press,Princeton, N.J., 1966.[35] C. Reutenauer, On symmetric functions related to Witt vectors and the free Liealgebra.
Adv. Math. (1995), 234–246.[36] K. Rietsch, Totally positive Toeplitz matrices and quantum cohomology of partialflag varieties.
J. Amer. Math. Soc. (2003), 363–392.[37] The Sage-Combinat community, Sage-Combinat: enhancing Sage asa toolbox for computer exploration in algebraic combinatorics, 2013. http://combinat.sagemath.org .[38] T. Scharf and J.-Y. Thibon, On Witt vectors and symmetric functions. AlgebraColloq. (1996), 231–238.[39] A. Stacey and S. Whitehouse, Tall–wraith monoids. arXiv:1102.3549v1.[40] W. A. Stein et al, Sage Mathematics Software (Version 5.4) . The Sage DevelopmentTeam, 2013. .[41] D. O. Tall and G. C. Wraith, Representable functors and operations on rings.
Proc.London Math. Soc. (3) (1970), 619–643.itt vectors, semirings, and total positivity [42] E. Thoma, Die unzerlegbaren, positiv-definiten Klassenfunktionen der abz¨ahlbarunendlichen, symmetrischen Gruppe. Math. Z. (1964), 40–61.[43] B. To¨en and M. Vaqui´e, Alg´ebrisation des vari´et´es analytiques complexes etcat´egories d´eriv´ees. Math. Ann. (2008), 789–831.[44] B. To¨en and M. Vaqui´e, Au-dessous de Spec Z . J. K-Theory (2009), 437–500.[45] C. Wilkerson, Lambda-rings, binomial domains, and vector bundles over C P ( ∞ ). Comm. Algebra (1982), 311–328.[46] E. Witt, Zyklische K¨orper und Algebren der Charakteristik p vom Grad p n . Struk-tur diskret bewerteter perfekter K¨orper mit vollkommenem Restklassen-k¨orper dercharakteristik p . J. Reine Angew. Math. (1937), 126–140.[47] E. Witt,
Collected papers. Gesammelte Abhandlungen . Springer-Verlag, Berlin, 1998.With an essay by G¨unter Harder on Witt vectors, Edited and with a preface inEnglish and German by Ina Kersten. J. Borgeritt vectors, semirings, and total positivity329