Introduction
Novikov algebras were introduced by Gelfand and Dorfman in 1979 in the study of Hamiltonian operators [8 ] , and independently by Balinsky and Novikov in 1985 in connection with Poisson brackets of hydrodynamic type [3 ] . Osborn [13 , 14 ] introduced the term Novikov algebra and studied the structure and classification of finite-dimensional simple Novikov algebras. In characteristic zero, Novikov algebras give rise to infinite-dimensional Lie algebras by affinization, and to Lie conformal algebras and vertex algebras [3 , 2 ] . Novikov bialgebras similarly yield infinite-dimensional Lie bialgebras and Lie conformal bialgebras [9 , 10 ] . A Novikov algebra is solvable if and only if it is right nilpotent [21 Theorem 3.3] . Recent results concern polynomial identities [6 ] and the classification of finite simple Novikov conformal algebras [11 ] .
Zelmanov [26 ] proved that every finite-dimensional simple Novikov algebra over an algebraically closed field of characteristic zero is one-dimensional. Every finite-dimensional simple Novikov superalgebra over a field of characteristic different from 2 has zero odd part [12 ] . In positive characteristic, Osborn proposed a classification of finite-dimensional simple Novikov algebras containing a nonzero idempotent [14 ] . Xu proposed the same classification without the assumption of a nonzero idempotent [23 Main Theorem 1] and classified the finite-dimensional irreducible bimodules over the algebras in his classification [23 Main Theorem 2] (see also [15 ] ). The classification of Osborn and Xu is incomplete in dimensions p n p^{n} p n with n ≥ 2 n \ge2 n ≥ 2 ; Remark 4.9 gives a p 2 p^{2} p 2 -dimensional simple Novikov algebra outside their family.
For a commutative associative algebra A A A , a derivation ∂ ∈ Der ( A ) \partial\in\operatorname{Der}(A) ∂ ∈ Der ( A ) , and a ∈ A a \in A a ∈ A , the twisted Gelfand–Dorfman construction gives the Novikov algebra A ( ∂ , a ) A(\partial,a) A ( ∂ , a ) with product
x ∘ y = x ( ∂ ( y ) + a y ) . x \circ y = x(\partial(y) + ay). x ∘ y = x ( ∂ ( y ) + a y ) .
This product is obtained by adding a x y axy a x y to the differential product x ∂ ( y ) x\partial(y) x ∂ ( y ) in [8 ] . The scalar twist was introduced in [7 ] , and the general twist a ∈ A a \in A a ∈ A in [24 , 25 ] . A related PBW embedding theorem for Novikov–Poisson algebras with a unital associative product was proved in [5 Theorem 3.3] .
By the structure theorem of Zhelyabin and Zakharov [27 Theorem 3 and Corollary 1] , every finite-dimensional simple Novikov algebra over an algebraically closed field of characteristic p > 2 p > 2 p > 2 is obtained by the twisted Gelfand–Dorfman construction, with A A A a ∂ \partial ∂ -simple truncated polynomial algebra. The one-dimensional case is included by taking ∂ = 0 \partial=0 ∂ = 0 and a = 1 a=1 a = 1 on the ground field. The corresponding existence theorem over arbitrary fields of positive characteristic was proved in [16 Theorem 2.4] , where automorphisms were also studied. An isomorphism criterion for these algebras was subsequently obtained in [18 ] . The remaining problem is to find explicit normal forms for these algebras and determine when they are isomorphic. The aim of this paper is to complete this classification.
In what follows, k k k is an algebraically closed field of characteristic p > 2 p>2 p > 2 , and F p \mathbb{F}_p F p denotes its prime subfield. For n ≥ 0 n\geq0 n ≥ 0 , let O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) be the truncated divided-power algebra with basis x ( 0 ) , … , x ( p n − 1 ) x^{(0)},\ldots,x^{(p^n-1)} x ( 0 ) , … , x ( p n − 1 ) and lowering derivation δ \delta δ . The definitions and conventions are collected in § 2.1 ; in particular, O ( 1 ; 0 ) = k \mathcal{O}(1;0)=k O ( 1 ; 0 ) = k and δ = 0 \delta=0 δ = 0 on O ( 1 ; 0 ) \mathcal{O}(1;0) O ( 1 ; 0 ) . For α = ( α 1 , … , α n ) ∈ k n \alpha=(\alpha_1,\ldots,\alpha_n)\in k^n α = ( α 1 , … , α n ) ∈ k n and β ∈ k \beta\in k β ∈ k , with β = 1 \beta=1 β = 1 if n = 0 n=0 n = 0 , put
b α , β = ∑ j = 1 n α j x ( p j − 2 ) + β x ( p n − 1 ) , N ( α , β ) = O ( 1 ; n ) ( δ , b α , β ) . (1) b_{\alpha,\beta}=\sum_{j=1}^{n}\alpha_jx^{(p^j-2)}+\beta x^{(p^n-1)},\qquad\mathcal{N}(\alpha,\beta)=\mathcal{O}(1;n)(\delta,b_{\alpha,\beta}). \tag*{(1)} b α , β = j = 1 ∑ n α j x ( p j − 2 ) + β x ( p n − 1 ) , N ( α , β ) = O ( 1 ; n ) ( δ , b α , β ) . (1)
When n = 0 n=0 n = 0 , the tuple α \alpha α is empty and N ( α , 1 ) = k \mathcal{N}(\alpha,1)=k N ( α , 1 ) = k with its usual product. In the basis y i = x ( i + 1 ) y_i=x^{(i+1)} y i = x ( i + 1 ) , − 1 ≤ i ≤ p n − 2 -1\leq i\leq p^n-2 − 1 ≤ i ≤ p n − 2 , the product of N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) is
y i ∘ y j = ( i + j + 1 j ) y i + j + β δ i , − 1 δ j , − 1 y p n − 2 + ∑ s = 1 n α s ( i + j + 2 i + 1 ) ( i + j + p s p s − 2 ) y i + j + p s − 1 . (2) y_i\circ y_j=\binom{i+j+1}{j}y_{i+j}+\beta\delta_{i,-1}\delta_{j,-1}y_{p^n-2}+\sum_{s=1}^{n}\alpha_s\binom{i+j+2}{i+1}\binom{i+j+p^s}{p^s-2}y_{i+j+p^s-1}. \tag*{(2)} y i ∘ y j = ( j i + j + 1 ) y i + j + β δ i , − 1 δ j , − 1 y p n − 2 + s = 1 ∑ n α s ( i + 1 i + j + 2 ) ( p s − 2 i + j + p s ) y i + j + p s − 1 . (2)
Here − 1 ≤ i , j ≤ p n − 2 -1\leq i,j\leq p^n-2 − 1 ≤ i , j ≤ p n − 2 , y r = 0 y_r=0 y r = 0 for r ∉ { − 1 , … , p n − 2 } r\notin\{-1,\ldots,p^n-2\} r ∈ / { − 1 , … , p n − 2 } , and δ i j \delta_{ij} δ ij is the Kronecker delta. By Lemma 2.7, each algebra N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) is simple of dimension p n p^n p n .
Theorem A (Theorem 4.4 below). (i) Every finite-dimensional simple Novikov algebra is isomorphic to N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) for some n ≥ 0 n\geq0 n ≥ 0 and parameters as above. (ii) Two algebras N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) and N ( α ′ , β ′ ) \mathcal{N}(\alpha',\beta') N ( α ′ , β ′ ) of dimension p n p^n p n are isomorphic if and only if there is c ∈ k × c\in k^\times c ∈ k × such that
α j ′ = c p j − 1 α j ( 1 ≤ j ≤ n ) , β ′ = c p n β . \alpha'_j=c^{p^j-1}\alpha_j\quad(1\leq j\leq n),\qquad\beta'=c^{p^n}\beta. α j ′ = c p j − 1 α j ( 1 ≤ j ≤ n ) , β ′ = c p n β .
The proof starts from the structure theorem in [27 Theorem 3 and Corollary 1] and the isomorphism criterion in [18 Theorem 3.3] . Using an argument from [19 ] , we reduce ∂ \partial ∂ to δ \delta δ by coordinate changes and multiplication by units. Successive coefficient elimination by admissible automorphisms with linear coefficient 1 1 1 gives a unique normal form b α , β b_{\alpha,\beta} b α , β . Scalings then give the parameter identifications in Theorem A. As a consequence, we classify finite-dimensional Novikov–Poisson algebras whose Novikov part is simple (Corollary 4.11 ).
For the definition and basic properties of Novikov bimodules, see [15 , 23 , 25 ] . A bimodule is called nontrivial if its two actions are not both zero. For each N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) and λ ∈ k \lambda\in k λ ∈ k , let M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) be the N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) -bimodule with basis v − 1 , … , v p n − 2 v_{-1},\ldots,v_{p^n-2} v − 1 , … , v p n − 2 and actions
y i ∘ v j = ( i + j + 1 j ) v i + j + λ ( i + j + 2 i + 1 ) v i + j + 1 , (3) y_i\circ v_j=\binom{i+j+1}{j}v_{i+j}+\lambda\binom{i+j+2}{i+1}v_{i+j+1}, \tag*{(3)} y i ∘ v j = ( j i + j + 1 ) v i + j + λ ( i + 1 i + j + 2 ) v i + j + 1 , (3)
v i ∘ y j = ( i + j + 1 j ) v i + j + β δ i , − 1 δ j , − 1 v p n − 2 + ∑ s = 1 n α s ( i + j + 2 i + 1 ) ( i + j + p s p s − 2 ) v i + j + p s − 1 . (4) v_i\circ y_j=\binom{i+j+1}{j}v_{i+j}+\beta\delta_{i,-1}\delta_{j,-1}v_{p^n-2}+\sum_{s=1}^{n}\alpha_s\binom{i+j+2}{i+1}\binom{i+j+p^s}{p^s-2}v_{i+j+p^s-1}. \tag*{(4)} v i ∘ y j = ( j i + j + 1 ) v i + j + β δ i , − 1 δ j , − 1 v p n − 2 + s = 1 ∑ n α s ( i + 1 i + j + 2 ) ( p s − 2 i + j + p s ) v i + j + p s − 1 . (4)
Here − 1 ≤ i , j ≤ p n − 2 -1\leq i,j\leq p^n-2 − 1 ≤ i , j ≤ p n − 2 , and v r = 0 v_r=0 v r = 0 for r ∉ { − 1 , … , p n − 2 } r\notin\{-1,\ldots,p^n-2\} r ∈ / { − 1 , … , p n − 2 } . For n = 0 n=0 n = 0 , these formulas give 1 ∘ v − 1 = λ v − 1 1\circ v_{-1}=\lambda v_{-1} 1 ∘ v − 1 = λ v − 1 and v − 1 ∘ 1 = v − 1 v_{-1}\circ1=v_{-1} v − 1 ∘ 1 = v − 1 . The parameter λ \lambda λ is the unique eigenvalue of the left action of 1 1 1 .
Theorem B. (Theorem 5.4 below). Fix N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) as in Theorem A. (i) For each λ ∈ k \lambda\in\mathbb{k} λ ∈ k , the bimodule M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) is nontrivial and irreducible of dimension p n p^n p n . Every nontrivial irreducible N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) -bimodule is isomorphic to one of these. (ii) The N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) -bimodules M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) and M ( α , β , λ ′ ) \mathcal{M}(\alpha,\beta,\lambda') M ( α , β , λ ′ ) are isomorphic if and only if λ = λ ′ \lambda= \lambda' λ = λ ′ .
For Theorem B, we use polynomial relations among right multiplication operators to recover an action of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) on every nontrivial irreducible bimodule, following [15 , 23 , 25 ] . After proving finite-dimensionality, we show that an eigenvector of the left action of 1 1 1 generates the bimodule freely over O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) . This gives M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) .
Section 2 recalls the twisted Gelfand–Dorfman construction and the isomorphism criterion. Section 3 normalizes the derivation. Sections 4 and 5 prove Theorems A and B, respectively.
Twisted Gelfand–Dorfman construction and isomorphisms
We recall divided-power algebras and the twisted Gelfand–Dorfman construction, then state the structure and isomorphism theorems.
Divided-power algebras
We use O ( 1 ) \mathcal{O}(1) O ( 1 ) and O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) for the one-variable divided-power algebras in [20 §3.2] . We write f ( r ) f^{(r)} f ( r ) for divided powers and f r f^r f r for ordinary powers. Binomial coefficients are computed in Z \mathbb{Z} Z and then viewed as elements of k \mathbb{k} k , with ( a b ) = 0 \binom{a}{b}=0 ( b a ) = 0 unless 0 ≤ b ≤ a 0 \le b \le a 0 ≤ b ≤ a .
Definition 2.1. The divided-power algebra in one variable is
O ( 1 ) = ⨁ r ≥ 0 k x ( r ) , x ( 0 ) = 1 , x = x ( 1 ) . \mathcal{O}(1)=\bigoplus_{r\ge0}\mathbb{k}x^{(r)},\qquad x^{(0)}=1,\qquad x=x^{(1)}. O ( 1 ) = r ≥ 0 ⨁ k x ( r ) , x ( 0 ) = 1 , x = x ( 1 ) .
with multiplication
x ( r ) x ( s ) = ( r + s r ) x ( r + s ) ( r , s ≥ 0 ) . (5) x^{(r)}x^{(s)}=\binom{r+s}{r}x^{(r+s)}\qquad(r,s\ge0). \tag*{(5)} x ( r ) x ( s ) = ( r r + s ) x ( r + s ) ( r , s ≥ 0 ) . (5)
For n ≥ 1 n\ge1 n ≥ 1 , the truncated divided-power algebra in one variable of height n n n is
O ( 1 ; n ) = ⨁ 0 ≤ r < p n k x ( r ) . (6) \mathcal{O}(1;n)=\bigoplus_{0\le r<p^n}\mathbb{k}x^{(r)}. \tag*{(6)} O ( 1 ; n ) = 0 ≤ r < p n ⨁ k x ( r ) . (6)
By convention, O ( 1 ; 0 ) = k \mathcal{O}(1;0)=\mathbb{k} O ( 1 ; 0 ) = k .
The product is commutative and associative; associativity follows from the integer identity
( r + s r ) ( r + s + t r + s ) = ( s + t s ) ( r + s + t r ) . \binom{r+s}{r}\binom{r+s+t}{r+s}
=
\binom{s+t}{s}\binom{r+s+t}{r}. ( r r + s ) ( r + s r + s + t ) = ( s s + t ) ( r r + s + t ) .
To see that (6) is a subalgebra, recall Lucas’s congruence
( a b ) = ∏ i ≥ 0 ( a i b i ) in k , (7) \binom{a}{b}=\prod_{i\ge0}\binom{a_i}{b_i}\quad\text{in }\mathbb{k}, \tag*{(7)} ( b a ) = i ≥ 0 ∏ ( b i a i ) in k , (7)
where a i , b i a_i,b_i a i , b i are the base-p p p digits of a , b ≥ 0 a,b\ge0 a , b ≥ 0 . If r , s < p n ≤ r + s r,s<p^n\le r+s r , s < p n ≤ r + s , addition in base p p p has a carry, so ( r + s r ) = 0 \binom{r+s}{r}=0 ( r r + s ) = 0 in k \mathbb{k} k . Thus O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is a unital subalgebra of O ( 1 ) \mathcal{O}(1) O ( 1 ) of dimension p n p^n p n , and products of its basis elements vanish when their indices add to at least p n p^n p n . We set x ( t ) = 0 x^{(t)}=0 x ( t ) = 0 for t < 0 t<0 t < 0 .
Following [19 §2.1] , write
O n = k [ X 1 , … , X n ] / ( X 1 p , … , X n p ) , O 0 = k , \mathcal{O}_n=\mathbb{k}[X_1,\ldots,X_n]/(X_1^p,\ldots,X_n^p),\qquad\mathcal{O}_0=\mathbb{k}, O n = k [ X 1 , … , X n ] / ( X 1 p , … , X n p ) , O 0 = k ,
and let x i x_i x i denote the residue class of X i X_i X i . For n > 0 n>0 n > 0 , Lucas’s congruence gives
( x ( p i − 1 ) ) j = j ! x ( j p i − 1 ) ( 1 ≤ i ≤ n , 0 ≤ j < p ) , ( x ( p i − 1 ) ) p = 0. (8) \left(x^{(p^{i-1})}\right)^j=j!x^{(jp^{i-1})}\qquad(1\le i\le n,\ 0\le j<p),\qquad
\left(x^{(p^{i-1})}\right)^p=0. \tag*{(8)} ( x ( p i − 1 ) ) j = j ! x ( j p i − 1 ) ( 1 ≤ i ≤ n , 0 ≤ j < p ) , ( x ( p i − 1 ) ) p = 0. (8)
Hence the assignment
O n → ∼ O ( 1 ; n ) , x i ⟼ x ( p i − 1 ) (9) \mathcal{O}_n\xrightarrow{\sim}\mathcal{O}(1;n),\qquad x_i\longmapsto x^{(p^{i-1})} \tag*{(9)} O n ∼ O ( 1 ; n ) , x i ⟼ x ( p i − 1 ) (9)
is an isomorphism of commutative associative algebras: it maps the monomial basis of O n \mathcal{O}_n O n to nonzero scalar multiples of the divided-power basis of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) . We identify these algebras, so that x i = x ( p i − 1 ) x_i=x^{(p^{i-1})} x i = x ( p i − 1 ) . In these polynomial coordinates,
x ( r ) = ∏ i = 0 n − 1 x i + 1 r i r i ! , r = ∑ i = 0 n − 1 r i p i , 0 ≤ r < p n , 0 ≤ r i < p . (10) x^{(r)}=\prod_{i=0}^{n-1}\frac{x_{i+1}^{r_i}}{r_i!},\qquad r=\sum_{i=0}^{n-1}r_ip^i,\qquad0\leq r<p^n,\qquad0\leq r_i<p. \tag*{(10)} x ( r ) = i = 0 ∏ n − 1 r i ! x i + 1 r i , r = i = 0 ∑ n − 1 r i p i , 0 ≤ r < p n , 0 ≤ r i < p . (10)
For 0 ≤ r ≤ n 0\leq r\leq n 0 ≤ r ≤ n , we identify O ( 1 ; r ) \mathcal{O}(1;r) O ( 1 ; r ) with the subalgebra of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) generated by x 1 , … , x r x_1,\ldots,x_r x 1 , … , x r .
For f = ∑ j a j x ( j ) f=\sum_j a_jx^{(j)} f = ∑ j a j x ( j ) , write [ x ( j ) ] f = a j [x^{(j)}]f=a_j [ x ( j ) ] f = a j and f ( 0 ) = a 0 f(0)=a_0 f ( 0 ) = a 0 . The algebra O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is local with maximal ideal
m = { f ∈ O ( 1 ; n ) : f ( 0 ) = 0 } = span k { x ( j ) : 1 ≤ j < p n } = ( x 1 , … , x n ) , \mathfrak{m}=\{f\in\mathcal{O}(1;n):f(0)=0\}=\operatorname{span}_{\mathbb{k}}\{x^{(j)}:1\leq j<p^n\}=(x_1,\ldots,x_n), m = { f ∈ O ( 1 ; n ) : f ( 0 ) = 0 } = span k { x ( j ) : 1 ≤ j < p n } = ( x 1 , … , x n ) ,
where m = 0 \mathfrak{m}=0 m = 0 if n = 0 n=0 n = 0 . The polynomial presentation and the Frobenius identity give
f p = f ( 0 ) p 1 , f ∈ O ( 1 ; n ) × ⟺ f ( 0 ) ≠ 0. (11) f^p=f(0)^p1,\qquad f\in\mathcal{O}(1;n)^\times\Longleftrightarrow f(0)\ne0. \tag*{(11)} f p = f ( 0 ) p 1 , f ∈ O ( 1 ; n ) × ⟺ f ( 0 ) = 0. (11)
Indeed, if f ( 0 ) = 0 f(0)=0 f ( 0 ) = 0 , then f f f is nilpotent; otherwise f − 1 = f ( 0 ) − p f p − 1 f^{-1}=f(0)^{-p}f^{p-1} f − 1 = f ( 0 ) − p f p − 1 .
The standard divided-power grading assigns degree j j j to x ( j ) x^{(j)} x ( j ) , so deg x i = p i − 1 \deg x_i=p^{i-1} deg x i = p i − 1 . For fixed n n n and every integer s ≥ 0 s\geq0 s ≥ 0 , put
F s = span k { x ( j ) : s ≤ j < p n } . (12) F^s=\operatorname{span}_{\mathbb{k}}\{x^{(j)}:s\leq j<p^n\}. \tag*{(12)} F s = span k { x ( j ) : s ≤ j < p n } . (12)
These subspaces form the standard divided-power filtration. We have F 0 = O ( 1 ; n ) F^0=\mathcal{O}(1;n) F 0 = O ( 1 ; n ) , F 1 = m F^1=\mathfrak{m} F 1 = m , F i F j ⊆ F i + j F^iF^j\subseteq F^{i+j} F i F j ⊆ F i + j for i , j ≥ 0 i,j\geq0 i , j ≥ 0 , and F s = 0 F^s=0 F s = 0 for s ≥ p n s\geq p^n s ≥ p n .
The lowering derivation on O ( 1 ) \mathcal{O}(1) O ( 1 ) is defined by
δ ( 1 ) = 0 , δ ( x ( r ) ) = x ( r − 1 ) ( r ≥ 1 ) . (13) \delta(1)=0,\qquad\delta(x^{(r)})=x^{(r-1)}\qquad(r\geq1). \tag*{(13)} δ ( 1 ) = 0 , δ ( x ( r ) ) = x ( r − 1 ) ( r ≥ 1 ) . (13)
Pascal’s identity proves the Leibniz rule: for r , s ≥ 0 r,s\geq0 r , s ≥ 0 with r + s > 0 r+s>0 r + s > 0 ,
δ ( x ( r ) ) x ( s ) + x ( r ) δ ( x ( s ) ) = ( ( r + s − 1 r − 1 ) + ( r + s − 1 r ) ) x ( r + s − 1 ) = δ ( x ( r ) x ( s ) ) . \delta(x^{(r)})x^{(s)}+x^{(r)}\delta(x^{(s)})=\left(\binom{r+s-1}{r-1}+\binom{r+s-1}{r}\right)x^{(r+s-1)}=\delta(x^{(r)}x^{(s)}). δ ( x ( r ) ) x ( s ) + x ( r ) δ ( x ( s ) ) = ( ( r − 1 r + s − 1 ) + ( r r + s − 1 ) ) x ( r + s − 1 ) = δ ( x ( r ) x ( s ) ) .
The derivation preserves every O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) , and we denote its restriction by the same symbol. In particular, δ = 0 \delta=0 δ = 0 on O ( 1 ; 0 ) \mathcal{O}(1;0) O ( 1 ; 0 ) . Its iterates satisfy
δ j ( x ( r ) ) = { x ( r − j ) , 0 ≤ j ≤ r , 0 , j > r . (14) \delta^j(x^{(r)})=
\begin{cases}
x^{(r-j)},&0\leq j\leq r,\\
0,&j>r.
\end{cases} \tag*{(14)} δ j ( x ( r ) ) = { x ( r − j ) , 0 , 0 ≤ j ≤ r , j > r . (14)
Thus ker δ = k 1 \ker\delta=\mathbb{k}1 ker δ = k 1 in O ( 1 ) \mathcal{O}(1) O ( 1 ) , and on O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) the operator δ \delta δ has a single nilpotent Jordan block of size p n p^n p n . For a derivation ∂ \partial ∂ of a commutative associative algebra A A A , we call A A A ∂ \partial ∂ -simple if it has no nonzero proper ∂ \partial ∂ -stable ideal. If a nonzero δ \delta δ -stable ideal of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) contains f = ∑ r = 0 s c r x ( r ) f=\sum_{r=0}^s c_rx^{(r)} f = ∑ r = 0 s c r x ( r ) with c s ≠ 0 c_s\ne0 c s = 0 , it contains δ s ( f ) = c s 1 \delta^s(f)=c_s1 δ s ( f ) = c s 1 . Hence O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is δ \delta δ -simple.
For n > 0 n>0 n > 0 , let ∂ i = ∂ / ∂ x i \partial_i=\partial/\partial x_i ∂ i = ∂ / ∂ x i be the coordinate derivations, so that ∂ i ( x j ) = δ i j \partial_i(x_j)=\delta_{ij} ∂ i ( x j ) = δ ij . The relations x i p = 0 x_i^p=0 x i p = 0 are preserved because p = 0 p=0 p = 0 in k \mathbb{k} k . Every derivation is uniquely determined by its values on the x i x_i x i , and therefore
Der ( O ( 1 ; n ) ) = ⨁ i = 1 n O ( 1 ; n ) ∂ i . \operatorname{Der}(\mathcal{O}(1;n))=\bigoplus_{i=1}^n\mathcal{O}(1;n)\partial_i. Der ( O ( 1 ; n )) = i = 1 ⨁ n O ( 1 ; n ) ∂ i .
Formula (10) and ( p − 1 ) ! = − 1 (p-1)!=-1 ( p − 1 )! = − 1 in k \mathbb{k} k give
δ ( x i ) = x ( p i − 1 − 1 ) = ( − 1 ) i − 1 ∏ j < i x j p − 1 , δ = ∑ i = 1 n ( − 1 ) i − 1 ( ∏ j < i x j p − 1 ) ∂ i . (15) \delta(x_i)=x^{(p^{i-1}-1)}=(-1)^{i-1}\prod_{j<i}x_j^{p-1},\qquad
\delta=\sum_{i=1}^n(-1)^{i-1}\left(\prod_{j<i}x_j^{p-1}\right)\partial_i. \tag*{(15)} δ ( x i ) = x ( p i − 1 − 1 ) = ( − 1 ) i − 1 j < i ∏ x j p − 1 , δ = i = 1 ∑ n ( − 1 ) i − 1 ( j < i ∏ x j p − 1 ) ∂ i . (15)
where an empty product is 1. Set Δ 0 = 0 \Delta_0=0 Δ 0 = 0 and, for 1 ≤ r ≤ n 1\leq r\leq n 1 ≤ r ≤ n , define
Δ r = ∂ 1 + x 1 p − 1 ∂ 2 + ⋯ + x 1 p − 1 ⋯ x r − 1 p − 1 ∂ r . (16) \Delta_r=\partial_1+x_1^{p-1}\partial_2+\cdots+x_1^{p-1}\cdots x_{r-1}^{p-1}\partial_r. \tag*{(16)} Δ r = ∂ 1 + x 1 p − 1 ∂ 2 + ⋯ + x 1 p − 1 ⋯ x r − 1 p − 1 ∂ r . (16)
This derivation acts trivially on x r + 1 , … , x n x_{r+1},\ldots,x_n x r + 1 , … , x n ; for r = n r=n r = n it is the Albert derivation [1 , 4 ] . Define τ r ∈ Aut ( O ( 1 ; r ) ) \tau_r \in\operatorname{Aut}(\mathcal{O}(1;r)) τ r ∈ Aut ( O ( 1 ; r )) by τ r ( x i ) = ( − 1 ) i − 1 x i \tau_r(x_i)=(-1)^{i-1}x_i τ r ( x i ) = ( − 1 ) i − 1 x i . Since p p p is odd, τ r ( x j p − 1 ) = x j p − 1 \tau_r(x_j^{p-1})=x_j^{p-1} τ r ( x j p − 1 ) = x j p − 1 , so on O ( 1 ; r ) \mathcal{O}(1;r) O ( 1 ; r )
τ r δ τ r − 1 = Δ r . (17) \tau_r\delta\tau_r^{-1}=\Delta_r. \tag*{(17)} τ r δ τ r − 1 = Δ r . (17)
Consequently, Δ r \Delta_r Δ r acts on O ( 1 ; r ) \mathcal{O}(1;r) O ( 1 ; r ) as a single nilpotent Jordan block of size p r p^r p r , and O ( 1 ; r ) \mathcal{O}(1;r) O ( 1 ; r ) is Δ r \Delta_r Δ r -simple.
The divided-power maps f ↦ f ( r ) f\mapsto f^{(r)} f ↦ f ( r ) on ⨁ m ≥ 1 k x ( m ) \bigoplus_{m\geq1} kx^{(m)} ⨁ m ≥ 1 k x ( m ) are determined by
( x ( m ) ) ( r ) = ( r m ) ! r ! ( m ! ) r x ( r m ) ( m ≥ 1 ) , \left(x^{(m)}\right)^{(r)}=\frac{(rm)!}{r!(m!)^r}x^{(rm)}\qquad(m\geq1), ( x ( m ) ) ( r ) = r ! ( m ! ) r ( r m )! x ( r m ) ( m ≥ 1 ) ,
( c f ) ( r ) = c r f ( r ) , ( f + g ) ( r ) = ∑ i = 0 r f ( i ) g ( r − i ) ( r ≥ 0 ) . (18) (cf)^{(r)}=c^r f^{(r)},\qquad(f+g)^{(r)}=\sum_{i=0}^{r}f^{(i)}g^{(r-i)}\qquad(r\geq0). \tag*{(18)} ( c f ) ( r ) = c r f ( r ) , ( f + g ) ( r ) = i = 0 ∑ r f ( i ) g ( r − i ) ( r ≥ 0 ) . (18)
Here f , g f,g f , g have zero constant coefficient and c ∈ k c\in k c ∈ k . The factorial quotient is an integer: it counts partitions of an r m rm r m -element set into r r r unordered blocks of size m m m . It is evaluated in Z \mathbb{Z} Z before reduction to k k k . The sum and scalar rules extend the monomial formula to every finite linear combination, with f ( 0 ) = 1 f^{(0)}=1 f ( 0 ) = 1 and f ( 1 ) = f f^{(1)}=f f ( 1 ) = f . For 0 ≤ r < p 0\leq r<p 0 ≤ r < p , this gives f ( r ) = f r / r ! f^{(r)}=f^r/r! f ( r ) = f r / r ! .
Lemma 2.2. For f , g ∈ O ( 1 ) f,g\in\mathcal{O}(1) f , g ∈ O ( 1 ) with zero constant coefficient and integers r , s ≥ 0 r,s\geq0 r , s ≥ 0 ,
f ( r ) f ( s ) = ( r + s r ) f ( r + s ) , ( f g ) ( s ) = f s g ( s ) , f^{(r)}f^{(s)}=\binom{r+s}{r}f^{(r+s)},\qquad(fg)^{(s)}=f^s g^{(s)}, f ( r ) f ( s ) = ( r r + s ) f ( r + s ) , ( f g ) ( s ) = f s g ( s ) , and
δ ( f ( r ) ) = δ ( f ) f ( r − 1 ) ( r ≥ 1 ) . (19) \delta\left(f^{(r)}\right)=\delta(f)f^{(r-1)}\qquad(r\geq1). \tag*{(19)} δ ( f ( r ) ) = δ ( f ) f ( r − 1 ) ( r ≥ 1 ) . (19)
Proof. Regard the coefficients of f f f and g g g as indeterminates over Z \mathbb{Z} Z . Over Q \mathbb{Q} Q , use x ( m ) = x m / m ! x^{(m)}=x^m/m! x ( m ) = x m / m ! and f ( r ) = f r / r ! f^{(r)}=f^r/r! f ( r ) = f r / r ! . All coefficients in the divided-power basis are integral polynomials in these indeterminates, so the identities remain valid over k k k . □ \square □
Remark 2.3. By (19) , δ \delta δ is a special derivation [20 §3.2] . We use Der ( O ( 1 ; n ) ) \operatorname{Der}(\mathcal{O}(1;n)) Der ( O ( 1 ; n )) for all derivations of the associative algebra. The divided powers f ( r ) f^{(r)} f ( r ) are computed in O ( 1 ) \mathcal{O}(1) O ( 1 ) , even when f ∈ O ( 1 ; n ) f\in\mathcal{O}(1;n) f ∈ O ( 1 ; n ) . They need not lie in O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) : for example, x ( p ) ≠ 0 x^{(p)}\ne0 x ( p ) = 0 in O ( 1 ) \mathcal{O}(1) O ( 1 ) whereas x p = 0 x^p=0 x p = 0 , and x ( p ) ∉ O ( 1 ; 1 ) x^{(p)}\notin\mathcal{O}(1;1) x ( p ) ∈ / O ( 1 ; 1 ) .
Simple Novikov algebras and their isomorphisms
Definition 2.4. (i) A Novikov algebra is a vector space N \mathcal{N} N with a bilinear product ∘ \circ ∘ satisfying, for all x , y , z ∈ N x,y,z\in\mathcal{N} x , y , z ∈ N ,
( x ∘ y ) ∘ z − x ∘ ( y ∘ z ) = ( y ∘ x ) ∘ z − y ∘ ( x ∘ z ) , (20) (x\circ y)\circ z-x\circ(y\circ z)=(y\circ x)\circ z-y\circ(x\circ z), \tag*{(20)} ( x ∘ y ) ∘ z − x ∘ ( y ∘ z ) = ( y ∘ x ) ∘ z − y ∘ ( x ∘ z ) , (20) ( x ∘ y ) ∘ z = ( x ∘ z ) ∘ y . (21) (x\circ y)\circ z=(x\circ z)\circ y. \tag*{(21)} ( x ∘ y ) ∘ z = ( x ∘ z ) ∘ y . (21) (ii) A Novikov algebra N \mathcal{N} N is simple if N ∘ N ≠ 0 \mathcal{N}\circ\mathcal{N}\ne0 N ∘ N = 0 and N \mathcal{N} N has no nonzero proper two-sided ideal.
(iii) For a Novikov algebra N \mathcal{N} N , the left and right multiplication operators are L x ( y ) = x ∘ y L_x(y)=x\circ y L x ( y ) = x ∘ y and R x ( y ) = y ∘ x R_x(y)=y\circ x R x ( y ) = y ∘ x .
By (21) , the right multiplications commute. Denote by the associative algebra
R ( N ) : = span k { R a 1 ⋯ R a r : r ≥ 1 , a i ∈ N } . \mathcal{R}(\mathcal{N}):=\operatorname{span}_k\{R_{a_1}\cdots R_{a_r}:r\geq1,\ a_i\in\mathcal{N}\}. R ( N ) := span k { R a 1 ⋯ R a r : r ≥ 1 , a i ∈ N } .
Definition 2.5. Let A \mathcal{A} A be a commutative associative algebra, ∂ ∈ Der ( A ) \partial\in\operatorname{Der}(\mathcal{A}) ∂ ∈ Der ( A ) , and a ∈ A a\in\mathcal{A} a ∈ A . The twisted Gelfand–Dorfman construction [24 , 25 ] assigns to ( A , ∂ , a ) (\mathcal{A},\partial,a) ( A , ∂ , a ) the Novikov algebra A ( ∂ , a ) \mathcal{A}(\partial,a) A ( ∂ , a ) on the vector space A \mathcal{A} A with product
x ∘ y = x ( ∂ ( y ) + a y ) ( x , y ∈ A ) . (22) x\circ y=x(\partial(y)+ay)\qquad(x,y\in\mathcal{A}). \tag*{(22)} x ∘ y = x ( ∂ ( y ) + a y ) ( x , y ∈ A ) . (22) For a = 0 a=0 a = 0 , this is the Gelfand–Dorfman construction [8 ] .
Theorem 2.6 (Zhelyabin and Zakharov [27 Theorem 3, Corollary 1] ). Every finite-dimensional simple Novikov algebra N \mathcal{N} N is isomorphic to O ( 1 ; n ) ( ∂ , a ) \mathcal{O}(1;n)(\partial,a) O ( 1 ; n ) ( ∂ , a ) for some n ≥ 0 n \ge0 n ≥ 0 , ∂ ∈ Der ( O ( 1 ; n ) ) \partial\in\operatorname{Der}(\mathcal{O}(1;n)) ∂ ∈ Der ( O ( 1 ; n )) , and a ∈ O ( 1 ; n ) a \in\mathcal{O}(1;n) a ∈ O ( 1 ; n ) , where O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is ∂ \partial ∂ -simple. The associative algebra R ( N ) \mathcal{R}(\mathcal{N}) R ( N ) is also isomorphic to O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) .
Lemma 2.7. Let n ≥ 0 n \ge0 n ≥ 0 and b ∈ A = O ( 1 ; n ) b \in\mathcal{A} = \mathcal{O}(1;n) b ∈ A = O ( 1 ; n ) , with b = 1 b = 1 b = 1 if n = 0 n = 0 n = 0 . The algebra N = A ( δ , b ) \mathcal{N} = \mathcal{A}(\delta,b) N = A ( δ , b ) is simple, and R ( N ) \mathcal{R}(\mathcal{N}) R ( N ) consists of all operators m c : x ↦ c x m_c : x \mapsto cx m c : x ↦ c x , c ∈ A c \in\mathcal{A} c ∈ A .
Proof. For n = 0 n = 0 n = 0 , the algebra is k \mathbb{k} k and R 1 = Id R_1 = \operatorname{Id} R 1 = Id . Suppose n > 0 n > 0 n > 0 , and put q = p n q = p^n q = p n and D = δ + m b D = \delta+ m_b D = δ + m b . Since R y = m D ( y ) R_y = m_{D(y)} R y = m D ( y ) and m c m d = m c d m_c m_d = m_{cd} m c m d = m c d , for the last assertion, it suffices to show that D ( A ) = span { D ( x ) ∣ x ∈ A } D(\mathcal{A}) = \operatorname{span}\{D(x) \mid x \in\mathcal{A}\} D ( A ) = span { D ( x ) ∣ x ∈ A } generates A \mathcal{A} A as an associative algebra. For 1 ≤ r < q 1 \le r < q 1 ≤ r < q , the element
D ( x ( r ) ) = x ( r − 1 ) + b x ( r ) D(x^{(r)}) = x^{(r-1)} + bx^{(r)} D ( x ( r ) ) = x ( r − 1 ) + b x ( r )
has lowest-degree term x ( r − 1 ) x^{(r-1)} x ( r − 1 ) . To obtain the highest-degree element, we use q ≥ 3 q \ge3 q ≥ 3 and
D ( x ( 2 ) ) D ( x ( q − 1 ) ) = x x ( q − 2 ) = − x ( q − 1 ) , D(x^{(2)})D(x^{(q-1)}) = x x^{(q-2)} = -x^{(q-1)}, D ( x ( 2 ) ) D ( x ( q − 1 ) ) = x x ( q − 2 ) = − x ( q − 1 ) ,
since all terms containing b b b have degree at least q q q and vanish, and ( q − 1 1 ) = − 1 \binom{q-1}{1} = -1 ( 1 q − 1 ) = − 1 in k \mathbb{k} k . These q q q elements have distinct lowest degrees, so they form a basis of A \mathcal{A} A .
Let I I I be a Novikov ideal of N \mathcal{N} N . It is stable under every m c m_c m c , so it is an associative ideal of A \mathcal{A} A . It is stable under D = L 1 D = L_1 D = L 1 and therefore under δ = D − m b \delta= D - m_b δ = D − m b . Since A \mathcal{A} A is δ \delta δ -simple, I = 0 I = 0 I = 0 or I = A I = \mathcal{A} I = A . Finally, 1 ∘ x = 1 + b x ≠ 0 1 \circ x = 1 + bx \ne0 1 ∘ x = 1 + b x = 0 , so N \mathcal{N} N is simple. □
Theorem 2.8 (Pozhidaev, Zakharov and Zhelyabin [18 Theorem 3.3, Corollary 3.1] ). Let A ( ∂ , a ) \mathcal{A}(\partial,a) A ( ∂ , a ) and A ′ ( ∂ ′ , a ′ ) \mathcal{A}'(\partial',a') A ′ ( ∂ ′ , a ′ ) be finite-dimensional simple Novikov algebras, where A \mathcal{A} A and A ′ \mathcal{A}' A ′ are unital commutative associative algebras. Every Novikov algebra isomorphism ϕ : A ( ∂ , a ) → A ′ ( ∂ ′ , a ′ ) \phi: \mathcal{A}(\partial,a) \to\mathcal{A}'(\partial',a') ϕ : A ( ∂ , a ) → A ′ ( ∂ ′ , a ′ ) has a unique expression
ϕ ( x ) = h − 1 σ ( x ) ( x ∈ A ) , (23) \phi(x) = h^{-1}\sigma(x) \qquad(x \in\mathcal{A}), \tag*{(23)} ϕ ( x ) = h − 1 σ ( x ) ( x ∈ A ) , (23) where σ : A → A ′ \sigma: \mathcal{A} \to\mathcal{A}' σ : A → A ′ is an associative algebra isomorphism and h ∈ ( A ′ ) × h \in(\mathcal{A}')^\times h ∈ ( A ′ ) × . With ∂ σ = σ ∂ σ − 1 \partial^\sigma= \sigma\partial\sigma^{-1} ∂ σ = σ ∂ σ − 1 , the data satisfy
∂ ′ = h ∂ σ , a ′ = ∂ σ ( h ) + h σ ( a ) . (24) \partial' = h\partial^\sigma, \qquad a' = \partial^\sigma(h) + h\sigma(a). \tag*{(24)} ∂ ′ = h ∂ σ , a ′ = ∂ σ ( h ) + hσ ( a ) . (24) Conversely, any σ \sigma σ and h h h satisfying (24) give an isomorphism by (23) .
Normalization of derivations
We reduce the algebras in Theorem 2.6 to O ( 1 ; n ) ( δ , b ) \mathcal{O}(1;n)(\delta,b) O ( 1 ; n ) ( δ , b ) by coordinate changes and multiplication by units. We use the polynomial coordinates x i x_i x i and the derivations ∂ i \partial_i ∂ i , Δ r \Delta_r Δ r from § 2.1 .
For ∂ ∈ Der ( O ( 1 ; n ) ) \partial\in\operatorname{Der}(\mathcal{O}(1;n)) ∂ ∈ Der ( O ( 1 ; n )) , write
∂ = ∂ s s + ∂ n i l , [ ∂ s s , ∂ n i l ] = 0 \partial= \partial_{\mathrm{ss}} + \partial_{\mathrm{nil}}, \qquad[\partial_{\mathrm{ss}},\partial_{\mathrm{nil}}] = 0 ∂ = ∂ ss + ∂ nil , [ ∂ ss , ∂ nil ] = 0
for its Jordan decomposition as a linear operator. Both components are derivations. Indeed, the Leibniz rule gives
( ∂ − ( λ + μ ) Id ) m ( f g ) = ∑ i = 0 m ( m i ) ( ∂ − λ Id ) i ( f ) ( ∂ − μ Id ) m − i ( g ) . \bigl(\partial- (\lambda+ \mu)\operatorname{Id}\bigr)^m(fg) = \sum_{i=0}^{m} \binom{m}{i}(\partial- \lambda\operatorname{Id})^i(f)(\partial- \mu\operatorname{Id})^{m-i}(g). ( ∂ − ( λ + μ ) Id ) m ( f g ) = i = 0 ∑ m ( i m ) ( ∂ − λ Id ) i ( f ) ( ∂ − μ Id ) m − i ( g ) .
If ( ∂ − λ Id ) s ( f ) = 0 (\partial- \lambda\operatorname{Id})^s(f) = 0 ( ∂ − λ Id ) s ( f ) = 0 and ( ∂ − μ Id ) t ( g ) = 0 (\partial- \mu\operatorname{Id})^t(g) = 0 ( ∂ − μ Id ) t ( g ) = 0 , the right-hand side vanishes for m ≥ s + t − 1 m \ge s + t - 1 m ≥ s + t − 1 . Thus the product of generalized eigenvectors of eigenvalues λ , μ \lambda,\mu λ , μ lies in the generalized eigenspace of λ + μ \lambda+ \mu λ + μ , and
∂ s s ( f g ) = ( λ + μ ) f g = ∂ s s ( f ) g + f ∂ s s ( g ) . \partial_{\mathrm{ss}}(fg) = (\lambda+ \mu)fg = \partial_{\mathrm{ss}}(f)g + f\partial_{\mathrm{ss}}(g). ∂ ss ( f g ) = ( λ + μ ) f g = ∂ ss ( f ) g + f ∂ ss ( g ) .
The assertion for ∂ n i l = ∂ − ∂ s s \partial_{\mathrm{nil}} = \partial- \partial_{\mathrm{ss}} ∂ nil = ∂ − ∂ ss follows.
For ( p , n ) ≠ ( 3 , 1 ) (p,n) \ne(3,1) ( p , n ) = ( 3 , 1 ) , the next normal form follows from [19 Theorem 2(ii)–(iii)] . We give a direct proof for all p > 2 p > 2 p > 2 .
Lemma 3.1. For n > 0 n>0 n > 0 , let ∂ ∈ Der ( O ( 1 ; n ) ) \partial\in\operatorname{Der}(\mathcal{O}(1;n)) ∂ ∈ Der ( O ( 1 ; n )) be such that O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is ∂ \partial ∂ -simple. Then there exist an integer r r r with 0 ≤ r ≤ n 0\le r\le n 0 ≤ r ≤ n , an automorphism σ ∈ Aut ( O ( 1 ; n ) ) \sigma\in\operatorname{Aut}(\mathcal{O}(1;n)) σ ∈ Aut ( O ( 1 ; n )) , and elements ν r + 1 , … , ν n ∈ k \nu_{r+1},\ldots,\nu_n\in\mathbb{k} ν r + 1 , … , ν n ∈ k linearly independent over F p \mathbb{F}_p F p such that
σ ∂ σ − 1 = Δ r + ∑ i = r + 1 n ν i ( 1 + x i ) ∂ i . (25) \sigma\partial\sigma^{-1}=\Delta_r+\sum_{i=r+1}^{n}\nu_i(1+x_i)\partial_i. \tag*{(25)} σ ∂ σ − 1 = Δ r + i = r + 1 ∑ n ν i ( 1 + x i ) ∂ i . (25) The integer r r r is determined by dim ker ∂ s s = p r \dim\ker\partial_{\mathrm{ss}}=p^r dim ker ∂ ss = p r .
Proof. Write A = O ( 1 ; n ) \mathcal{A}=\mathcal{O}(1;n) A = O ( 1 ; n ) . Differential simplicity first determines the kernel of ∂ \partial ∂ : if f ∈ ker ∂ f\in\ker\partial f ∈ ker ∂ , then A ( f − f ( 0 ) ) ⊆ m \mathcal{A}(f-f(0))\subseteq\mathfrak{m} A ( f − f ( 0 )) ⊆ m is a proper ∂ \partial ∂ -stable ideal. Hence f = f ( 0 ) f=f(0) f = f ( 0 ) and ker ∂ = k 1 \ker\partial=\mathbb{k}1 ker ∂ = k 1 . Let Λ \Lambda Λ be the spectrum of ∂ \partial ∂ , and put
A λ = ker ( ∂ − λ Id ) p n , A = ⨁ λ ∈ Λ A λ . \mathcal{A}_{\lambda}=\ker(\partial-\lambda\operatorname{Id})^{p^n},\qquad
\mathcal{A}=\bigoplus_{\lambda\in\Lambda}\mathcal{A}_{\lambda}. A λ = ker ( ∂ − λ Id ) p n , A = λ ∈ Λ ⨁ A λ .
In particular, A 0 = ker ∂ s s \mathcal{A}_0=\ker\partial_{\mathrm{ss}} A 0 = ker ∂ ss is a unital ∂ \partial ∂ -stable subalgebra.
To compare the generalized eigenspaces with A 0 \mathcal{A}_0 A 0 , we use the fact that every nonzero eigenvector is a unit. Indeed, if 0 ≠ w ∈ ker ( ∂ − λ Id ) 0\ne w\in\ker(\partial-\lambda\operatorname{Id}) 0 = w ∈ ker ( ∂ − λ Id ) , then
∂ ( a w ) = ( ∂ ( a ) + λ a ) w ( a ∈ A ) . \partial(aw)=(\partial(a)+\lambda a)w\qquad(a\in\mathcal{A}). ∂ ( a w ) = ( ∂ ( a ) + λa ) w ( a ∈ A ) .
Thus A w \mathcal{A}w A w is a nonzero ∂ \partial ∂ -stable ideal, so A w = A \mathcal{A}w=\mathcal{A} A w = A and w w w is a unit. We may therefore choose eigenvectors z λ z_\lambda z λ with z λ ( 0 ) = 1 z_\lambda(0)=1 z λ ( 0 ) = 1 . Then z λ p = 1 z_\lambda^p=1 z λ p = 1 and
( ∂ − λ Id ) j ( z λ f ) = z λ ∂ j ( f ) ( f ∈ A , j ≥ 0 ) . (\partial-\lambda\operatorname{Id})^j(z_\lambda f)=z_\lambda\partial^j(f)\qquad(f\in\mathcal{A},\ j\ge0). ( ∂ − λ Id ) j ( z λ f ) = z λ ∂ j ( f ) ( f ∈ A , j ≥ 0 ) .
Since z λ z_\lambda z λ is a unit, this gives A λ = z λ A 0 \mathcal{A}_\lambda=z_\lambda\mathcal{A}_0 A λ = z λ A 0 . Moreover,
∂ ( z λ z μ ) = ( λ + μ ) z λ z μ , ∂ ( z λ − 1 ) = − λ z λ − 1 . \partial(z_\lambda z_\mu)=(\lambda+\mu)z_\lambda z_\mu,\qquad
\partial(z_\lambda^{-1})=-\lambda z_\lambda^{-1}. ∂ ( z λ z μ ) = ( λ + μ ) z λ z μ , ∂ ( z λ − 1 ) = − λ z λ − 1 .
The products and inverses are nonzero, so Λ \Lambda Λ is a finite F p \mathbb{F}_p F p -subspace of k \mathbb{k} k . Consequently, for some 0 ≤ r ≤ n 0\le r\le n 0 ≤ r ≤ n ,
p n = ∣ Λ ∣ dim A 0 , dim A 0 = p r , dim F p Λ = n − r . p^n=|\Lambda|\dim\mathcal{A}_0,\qquad
\dim\mathcal{A}_0=p^r,\qquad
\dim_{\mathbb{F}_p}\Lambda=n-r. p n = ∣Λ∣ dim A 0 , dim A 0 = p r , dim F p Λ = n − r .
Choose an F p \mathbb{F}_p F p -basis ν r + 1 , … , ν n \nu_{r+1},\ldots,\nu_n ν r + 1 , … , ν n of Λ \Lambda Λ and eigenvectors z i = z ν i z_i=z_{\nu_i} z i = z ν i . For 0 ≤ a i < p 0\le a_i<p 0 ≤ a i < p ,
∂ ( ∏ i = r + 1 n z i a i ) = ( ∑ i = r + 1 n a i ν i ) ∏ i = r + 1 n z i a i . \partial\left(\prod_{i=r+1}^{n}z_i^{a_i}\right)
=\left(\sum_{i=r+1}^{n}a_i\nu_i\right)\prod_{i=r+1}^{n}z_i^{a_i}. ∂ ( i = r + 1 ∏ n z i a i ) = ( i = r + 1 ∑ n a i ν i ) i = r + 1 ∏ n z i a i .
These eigenvalues exhaust Λ \Lambda Λ without repetition, so the generalized eigenspace decomposition takes the form
A = ⨁ 0 ≤ a r + 1 , … , a n < p A 0 ∏ i = r + 1 n z i a i . \mathcal{A}=\bigoplus_{0\le a_{r+1},\ldots,a_n<p}\mathcal{A}_0\prod_{i=r+1}^{n}z_i^{a_i}. A = 0 ≤ a r + 1 , … , a n < p ⨁ A 0 i = r + 1 ∏ n z i a i .
This decomposition reduces the construction of coordinates to the subalgebra A 0 \mathcal{A}_0 A 0 . The restriction ∂ 0 = ∂ ∣ A 0 \partial_0=\partial|_{\mathcal{A}_0} ∂ 0 = ∂ ∣ A 0 is nilpotent with one-dimensional kernel, so it has a single Jordan block and
∂ 0 p r = 0 , ker ∂ 0 = k 1 , im ∂ 0 = ker ∂ 0 p r − 1 . \partial_0^{p^r}=0,\qquad
\ker\partial_0=\mathbb{k}1,\qquad
\operatorname{im}\partial_0=\ker\partial_0^{p^r-1}. ∂ 0 p r = 0 , ker ∂ 0 = k 1 , im ∂ 0 = ker ∂ 0 p r − 1 .
Starting with ψ 0 : k ↪ A 0 \psi_0:\mathbb{k}\hookrightarrow\mathcal{A}_0 ψ 0 : k ↪ A 0 , we construct injective unital algebra homomorphisms
ψ j : O ( 1 ; j ) ⟶ A 0 , x i ⟼ u i ∈ A 0 ∩ m , ∂ 0 ψ j = ψ j Δ j . \psi_j:\mathcal{O}(1;j)\longrightarrow\mathcal{A}_0,\qquad
x_i\longmapsto u_i\in\mathcal{A}_0\cap\mathfrak{m},\qquad
\partial_0\psi_j=\psi_j\Delta_j. ψ j : O ( 1 ; j ) ⟶ A 0 , x i ⟼ u i ∈ A 0 ∩ m , ∂ 0 ψ j = ψ j Δ j .
Suppose ψ j \psi_j ψ j has been constructed and j < r j<r j < r . For the next coordinate, we need a preimage under ∂ 0 \partial_0 ∂ 0 of t j = ∏ i = 1 j u i p − 1 t_j=\prod_{i=1}^{j}u_i^{p-1} t j = ∏ i = 1 j u i p − 1 , with t 0 = 1 t_0=1 t 0 = 1 . Since Δ j p j = 0 \Delta_j^{p^j}=0 Δ j p j = 0 on O ( 1 ; j ) \mathcal{O}(1;j) O ( 1 ; j ) ,
∂ 0 p j ( t j ) = ψ j ( Δ j p j ( ∏ i = 1 j x i p − 1 ) ) = 0. \partial_0^{p^j}(t_j)=\psi_j\left(\Delta_j^{p^j}\left(\prod_{i=1}^{j}x_i^{p-1}\right)\right)=0. ∂ 0 p j ( t j ) = ψ j ( Δ j p j ( i = 1 ∏ j x i p − 1 ) ) = 0.
As p j ≤ p r − 1 p^j \leq p^r-1 p j ≤ p r − 1 , we obtain
t j ∈ ker ∂ 0 p j ⊆ ker ∂ 0 p r − 1 = im ∂ 0 . t_j \in\ker\partial_0^{p^j} \subseteq\ker\partial_0^{p^r-1} = \operatorname{im}\partial_0. t j ∈ ker ∂ 0 p j ⊆ ker ∂ 0 p r − 1 = im ∂ 0 .
Choose v ∈ A 0 v \in\mathcal{A}_0 v ∈ A 0 with ∂ 0 ( v ) = t j \partial_0(v)=t_j ∂ 0 ( v ) = t j and subtract its constant term to obtain
u j + 1 = v − v ( 0 ) ∈ A 0 ∩ m , u j + 1 p = 0 , ∂ 0 ( u j + 1 ) = t j . u_{j+1}=v-v(0)\in\mathcal{A}_0\cap\mathfrak{m},\qquad u_{j+1}^p=0,\qquad\partial_0(u_{j+1})=t_j. u j + 1 = v − v ( 0 ) ∈ A 0 ∩ m , u j + 1 p = 0 , ∂ 0 ( u j + 1 ) = t j .
Thus x j + 1 ↦ u j + 1 x_{j+1}\mapsto u_{j+1} x j + 1 ↦ u j + 1 extends ψ j \psi_j ψ j to a unital algebra homomorphism ψ j + 1 \psi_{j+1} ψ j + 1 satisfying ∂ ψ j + 1 = ψ j + 1 Δ j + 1 \partial\psi_{j+1}=\psi_{j+1}\Delta_{j+1} ∂ ψ j + 1 = ψ j + 1 Δ j + 1 . Its kernel is a proper Δ j + 1 \Delta_{j+1} Δ j + 1 -stable ideal, so it is zero by § 2.1 . This completes the induction. Since dim O ( 1 ; r ) = dim A 0 = p r \dim\mathcal{O}(1;r)=\dim\mathcal{A}_0=p^r dim O ( 1 ; r ) = dim A 0 = p r , we have A 0 = k [ u 1 , … , u r ] \mathcal{A}_0=\mathbb{k}[u_1,\ldots,u_r] A 0 = k [ u 1 , … , u r ] .
The remaining coordinates come from the chosen eigenvectors: for i > r i>r i > r , put u i = z i − 1 u_i=z_i-1 u i = z i − 1 . Then
u i p = z i p − 1 = 0 , ∂ ( u i ) = ν i ( 1 + u i ) . u_i^p=z_i^p-1=0,\qquad\partial(u_i)=\nu_i(1+u_i). u i p = z i p − 1 = 0 , ∂ ( u i ) = ν i ( 1 + u i ) .
The preceding decomposition gives A = k [ u 1 , … , u n ] \mathcal{A}=\mathbb{k}[u_1,\ldots,u_n] A = k [ u 1 , … , u n ] . Hence ϕ ( x i ) = u i \phi(x_i)=u_i ϕ ( x i ) = u i defines a surjective algebra endomorphism of A \mathcal{A} A , which is an automorphism by finite-dimensionality. On the generators,
ϕ − 1 ∂ ϕ ( x i ) = { ∏ ℓ = 1 i − 1 x ℓ p − 1 , 1 ≤ i ≤ r , ν i ( 1 + x i ) , r < i ≤ n . \phi^{-1}\partial\phi(x_i)=
\begin{cases}
\prod_{\ell=1}^{i-1}x_\ell^{p-1}, & 1\leq i\leq r,\\
\nu_i(1+x_i), & r<i\leq n.
\end{cases} ϕ − 1 ∂ ϕ ( x i ) = { ∏ ℓ = 1 i − 1 x ℓ p − 1 , ν i ( 1 + x i ) , 1 ≤ i ≤ r , r < i ≤ n .
Thus σ = ϕ − 1 \sigma=\phi^{-1} σ = ϕ − 1 gives (25) . □
Corollary 3.2. Let n > 0 n>0 n > 0 and ∂ ∈ Der ( O ( 1 ; n ) ) \partial\in\operatorname{Der}(\mathcal{O}(1;n)) ∂ ∈ Der ( O ( 1 ; n )) . Then O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is ∂ \partial ∂ -simple if and only if there exist σ ∈ Aut ( O ( 1 ; n ) ) \sigma\in\operatorname{Aut}(\mathcal{O}(1;n)) σ ∈ Aut ( O ( 1 ; n )) and h ∈ O ( 1 ; n ) × h\in\mathcal{O}(1;n)^\times h ∈ O ( 1 ; n ) × such that
h σ ∂ σ − 1 = Δ n . (26) h\sigma\partial\sigma^{-1}=\Delta_n. \tag*{(26)} hσ ∂ σ − 1 = Δ n . (26) Proof. Write A = O ( 1 ; n ) \mathcal{A}=\mathcal{O}(1;n) A = O ( 1 ; n ) and suppose that A \mathcal{A} A is ∂ \partial ∂ -simple. By Lemma 3.1 , after conjugation we may assume that
∂ = Δ r + ∑ i = r + 1 n ν i ( 1 + x i ) ∂ i . (27) \partial=\Delta_r+\sum_{i=r+1}^{n}\nu_i(1+x_i)\partial_i. \tag*{(27)} ∂ = Δ r + i = r + 1 ∑ n ν i ( 1 + x i ) ∂ i . (27) where ν r + 1 , … , ν n \nu_{r+1},\ldots,\nu_n ν r + 1 , … , ν n are linearly independent over F p \mathbb{F}_p F p . If r = n r=n r = n , then ∂ = Δ n \partial=\Delta_n ∂ = Δ n . Suppose r < n r<n r < n . We will use a unit multiple of ∂ \partial ∂ to increase r r r by one in the normal form. Put
u = 1 + x r + 1 , ν = ν r + 1 ≠ 0 , E = u ∂ . u=1+x_{r+1},\qquad\nu=\nu_{r+1}\ne0,\qquad E=u\partial. u = 1 + x r + 1 , ν = ν r + 1 = 0 , E = u ∂ . For every ideal I I I of A \mathcal{A} A , the fact that u u u is a unit gives
E ( I ) ⊆ I ⟺ ∂ ( I ) ⊆ I . E(I)\subseteq I\quad\Longleftrightarrow\quad\partial(I)\subseteq I. E ( I ) ⊆ I ⟺ ∂ ( I ) ⊆ I . Thus A \mathcal{A} A is E E E -simple. Since ∂ ( u ) = ν u \partial(u)=\nu u ∂ ( u ) = ν u and u p = 1 u^p=1 u p = 1 , we have
∂ m u = m u ( ∂ + ν Id ) , E s = m u s ∏ j = 0 s − 1 ( ∂ + j ν Id ) ( s ≥ 1 ) , \partial m_u=m_u(\partial+\nu\operatorname{Id}),\qquad
E^s=m_{u^s}\prod_{j=0}^{s-1}(\partial+j\nu\operatorname{Id})\quad(s\geq1), ∂ m u = m u ( ∂ + ν Id ) , E s = m u s j = 0 ∏ s − 1 ( ∂ + j ν Id ) ( s ≥ 1 ) , where the second identity follows by induction. Taking s = p s=p s = p gives
E p = ∏ c ∈ F p ( ∂ + c ν Id ) = ∂ p − ν p − 1 ∂ . (28) E^p=\prod_{c\in\mathbb{F}_p}(\partial+c\nu\operatorname{Id})=\partial^p-\nu^{p-1}\partial. \tag*{(28)} E p = c ∈ F p ∏ ( ∂ + c ν Id ) = ∂ p − ν p − 1 ∂ . (28) using ∏ c ∈ F p ( X + c ν ) = X p − ν p − 1 X \prod_{c\in\mathbb{F}_p}(X+c\nu)=X^p-\nu^{p-1}X ∏ c ∈ F p ( X + c ν ) = X p − ν p − 1 X .
To determine how this multiplication changes r r r , we compute the generalized zero eigenspace of E E E . By Lemma 3.1 and its proof,
Λ = span F p { ν r + 1 , … , ν n } , A = ⨁ λ ∈ Λ A λ , \Lambda=\operatorname{span}_{\mathbb{F}_p}\{\nu_{r+1},\ldots,\nu_n\},\qquad
\mathcal{A}=\bigoplus_{\lambda\in\Lambda}\mathcal{A}_\lambda, Λ = span F p { ν r + 1 , … , ν n } , A = λ ∈ Λ ⨁ A λ , and on each generalized eigenspace,
dim A λ = p r , ( ∂ − λ Id ) p r ∣ A λ = 0. \dim\mathcal{A}_\lambda=p^r,\qquad
(\partial-\lambda\operatorname{Id})^{p^r}\big|_{\mathcal{A}_\lambda}=0. dim A λ = p r , ( ∂ − λ Id ) p r A λ = 0.
Since r ≤ n − 1 r \le n-1 r ≤ n − 1 , equation (28) yields
E p n ∣ A λ = ( ∂ p − ν p − 1 ∂ ) p n − 1 ∣ A λ = ( λ p n − ν ( p − 1 ) p n − 1 λ p n − 1 ) Id = ( λ p − ν p − 1 λ ) p n − 1 Id . \begin{aligned}
E^{p^n}\big|_{\mathcal{A}_{\lambda}} &= \left(\partial^p-\nu^{p-1}\partial\right)^{p^{n-1}}\big|_{\mathcal{A}_{\lambda}} \\
&= \left(\lambda^{p^n}-\nu^{(p-1)p^{n-1}}\lambda^{p^{n-1}}\right)\operatorname{Id} \\
&= \left(\lambda^p-\nu^{p-1}\lambda\right)^{p^{n-1}}\operatorname{Id}.
\end{aligned} E p n A λ = ( ∂ p − ν p − 1 ∂ ) p n − 1 A λ = ( λ p n − ν ( p − 1 ) p n − 1 λ p n − 1 ) Id = ( λ p − ν p − 1 λ ) p n − 1 Id .
The scalar vanishes exactly when λ ∈ F p ν \lambda\in\mathbb{F}_{p^\nu} λ ∈ F p ν . Since dim A = p n \dim\mathcal{A}=p^n dim A = p n , we therefore obtain
ker E s s = ker E p n = ⨁ c ∈ F p A c ν , \ker E_{\mathrm{ss}}=\ker E^{p^n}=\bigoplus_{c\in\mathbb{F}_p}\mathcal{A}_{c\nu}, ker E ss = ker E p n = c ∈ F p ⨁ A c ν ,
dim ker E s s = p ⋅ p r = p r + 1 . (29) \dim\ker E_{\mathrm{ss}}=p\cdot p^r=p^{r+1}. \tag*{(29)} dim ker E ss = p ⋅ p r = p r + 1 . (29)
Applying Lemma 3.1 to E E E replaces r r r by r + 1 r+1 r + 1 in (27) . Repeating this step n − r n-r n − r times gives Δ n \Delta_n Δ n . The successive changes combine into a single change of the required form: for automorphisms σ , ρ \sigma,\rho σ , ρ and units h , g h,g h , g , we have
g ( h ∂ σ ) ρ = g ρ ( h ) ∂ ρ σ . g(h\partial^\sigma)^\rho=g\rho(h)\partial^{\rho\sigma}. g ( h ∂ σ ) ρ = g ρ ( h ) ∂ ρ σ .
and g ρ ( h ) g\rho(h) g ρ ( h ) is again a unit.
Conversely, if h ∂ σ = Δ n h\partial^\sigma=\Delta_n h ∂ σ = Δ n , then for every ideal I I I of A \mathcal{A} A ,
∂ ( I ) ⊆ I ⟺ Δ n ( σ ( I ) ) ⊆ σ ( I ) . \partial(I)\subseteq I\quad\Longleftrightarrow\quad\Delta_n(\sigma(I))\subseteq\sigma(I). ∂ ( I ) ⊆ I ⟺ Δ n ( σ ( I )) ⊆ σ ( I ) .
The Δ n \Delta_n Δ n -simplicity of A \mathcal{A} A gives I = 0 I=0 I = 0 or I = A I=\mathcal{A} I = A . □
Proposition 3.3. Every finite-dimensional simple Novikov algebra over k k k is isomorphic to
O ( 1 ; n ) ( δ , b ) , x ∘ y = x ( δ ( y ) + b y ) , (30) \mathcal{O}(1;n)(\delta,b),\qquad x\circ y=x\bigl(\delta(y)+by\bigr), \tag*{(30)} O ( 1 ; n ) ( δ , b ) , x ∘ y = x ( δ ( y ) + b y ) , (30) for some n ≥ 0 n\ge0 n ≥ 0 and b ∈ O ( 1 ; n ) b\in\mathcal{O}(1;n) b ∈ O ( 1 ; n ) , with b = 1 b=1 b = 1 if n = 0 n=0 n = 0 .
Proof. For N ≅ k \mathcal{N}\cong k N ≅ k , take n = 0 n=0 n = 0 and b = 1 b=1 b = 1 . By Theorem 2.6, it remains to consider O ( 1 ; n ) ( ∂ , a ) \mathcal{O}(1;n)(\partial,a) O ( 1 ; n ) ( ∂ , a ) with n > 0 n>0 n > 0 and O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) ∂ \partial ∂ -simple. Corollary 3.2 gives σ 0 ∈ Aut ( O ( 1 ; n ) ) \sigma_0\in\operatorname{Aut}(\mathcal{O}(1;n)) σ 0 ∈ Aut ( O ( 1 ; n )) and h 0 ∈ O ( 1 ; n ) × h_0\in\mathcal{O}(1;n)^\times h 0 ∈ O ( 1 ; n ) × with h 0 ∂ σ 0 = Δ n h_0\partial^{\sigma_0}=\Delta_n h 0 ∂ σ 0 = Δ n . Using τ n δ τ n − 1 = Δ n \tau_n\delta\tau_n^{-1}=\Delta_n τ n δ τ n − 1 = Δ n from (17) , put
σ = τ n − 1 σ 0 , h = τ n − 1 ( h 0 ) . \sigma=\tau_n^{-1}\sigma_0,\qquad h=\tau_n^{-1}(h_0). σ = τ n − 1 σ 0 , h = τ n − 1 ( h 0 ) .
Then
h ∂ σ = τ n − 1 ( h 0 ∂ σ 0 ) τ n = τ n − 1 Δ n τ n = δ . h\partial^\sigma=\tau_n^{-1}(h_0\partial^{\sigma_0})\tau_n=\tau_n^{-1}\Delta_n\tau_n=\delta. h ∂ σ = τ n − 1 ( h 0 ∂ σ 0 ) τ n = τ n − 1 Δ n τ n = δ .
Set
b = ∂ σ ( h ) + h σ ( a ) = h − 1 δ ( h ) + h σ ( a ) . (31) b=\partial^\sigma(h)+h\sigma(a)=h^{-1}\delta(h)+h\sigma(a). \tag*{(31)} b = ∂ σ ( h ) + hσ ( a ) = h − 1 δ ( h ) + hσ ( a ) . (31)
By Lemma 2.7, O ( 1 ; n ) ( δ , b ) \mathcal{O}(1;n)(\delta,b) O ( 1 ; n ) ( δ , b ) is simple. Theorem 2.8 now gives
h − 1 σ : O ( 1 ; n ) ( ∂ , a ) → ∼ O ( 1 ; n ) ( δ , b ) . h^{-1}\sigma:\mathcal{O}(1;n)(\partial,a)\xrightarrow{\sim}\mathcal{O}(1;n)(\delta,b). h − 1 σ : O ( 1 ; n ) ( ∂ , a ) ∼ O ( 1 ; n ) ( δ , b ) .
□
Classification of simple algebras
By Proposition 3.3, the classification of finite-dimensional simple Novikov algebras reduces to that of O ( 1 ; n ) ( δ , b ) \mathcal{O}(1;n)(\delta,b) O ( 1 ; n ) ( δ , b ) . We use admissible automorphisms to reduce b b b to b α , β b_{\alpha,\beta} b α , β and prove that all isomorphisms between the resulting normal forms arise from scalings. This proves Theorem A and also yields their automorphism groups and a classification of Novikov–Poisson algebras with simple Novikov part.
Normal forms
Let n > 0 n>0 n > 0 . By Lemma 2.7 and Theorem 2.8 , the isomorphisms among the algebras O ( 1 ; n ) ( δ , b ) \mathcal{O}(1;n)(\delta,b) O ( 1 ; n ) ( δ , b ) , b ∈ O ( 1 ; n ) b\in\mathcal{O}(1;n) b ∈ O ( 1 ; n ) , are the maps h − 1 σ h^{-1}\sigma h − 1 σ with ( σ , h ) ∈ G n (\sigma,h)\in G_n ( σ , h ) ∈ G n , where
G n = { ( σ , h ) ∈ Aut ( O ( 1 ; n ) ) × O ( 1 ; n ) × : h δ σ = δ } , δ σ = σ δ σ − 1 . (32) G_n=\left\{(\sigma,h)\in\operatorname{Aut}(\mathcal{O}(1;n))\times\mathcal{O}(1;n)^\times:h\delta^\sigma=\delta\right\},\qquad\delta^\sigma=\sigma\delta\sigma^{-1}. \tag*{(32)} G n = { ( σ , h ) ∈ Aut ( O ( 1 ; n )) × O ( 1 ; n ) × : h δ σ = δ } , δ σ = σ δ σ − 1 . (32)
This is a group under the multiplication
( τ , g ) ( σ , h ) = ( τ σ , g τ ( h ) ) . (\tau,g)(\sigma,h)=(\tau\sigma,g\tau(h)). ( τ , g ) ( σ , h ) = ( τ σ , g τ ( h )) .
Indeed,
( g − 1 τ ) ( h − 1 σ ) = ( g τ ( h ) ) − 1 τ σ . (g^{-1}\tau)(h^{-1}\sigma)=(g\tau(h))^{-1}\tau\sigma. ( g − 1 τ ) ( h − 1 σ ) = ( g τ ( h ) ) − 1 τ σ .
The corresponding action on O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is
( σ , h ) ∗ b = δ σ ( h ) + h σ ( b ) . (33) (\sigma,h)*b=\delta^\sigma(h)+h\sigma(b). \tag*{(33)} ( σ , h ) ∗ b = δ σ ( h ) + hσ ( b ) . (33)
An automorphism σ \sigma σ of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is admissible if σ ( x ( r ) ) = σ ( x ) ( r ) \sigma(x^{(r)})=\sigma(x)^{(r)} σ ( x ( r ) ) = σ ( x ) ( r ) in O ( 1 ) \mathcal{O}(1) O ( 1 ) for all 0 ≤ r < p n 0\leq r<p^n 0 ≤ r < p n . Wilson described these automorphisms for p ≥ 5 p\geq5 p ≥ 5 in [22 Lemmas 9–10 and Corollaries 1–2 to Lemma 9] . We prove the following criterion for p > 2 p>2 p > 2 .
Proposition 4.1. The map ( σ , h ) ↦ σ (\sigma,h)\mapsto\sigma ( σ , h ) ↦ σ is an isomorphism from G n G_n G n onto the group of admissible automorphisms of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) . The map σ ↦ σ ( x ) \sigma\mapsto\sigma(x) σ ↦ σ ( x ) is a bijection from this group onto the set of elements
f = ∑ r = 1 p n − 1 a r x ( r ) , a 1 ≠ 0 , a p j = 0 ( 1 ≤ j < n ) . (34) f=\sum_{r=1}^{p^n-1}a_rx^{(r)},\qquad a_1\ne0,\qquad a_{p^j}=0\quad(1\leq j<n). \tag*{(34)} f = r = 1 ∑ p n − 1 a r x ( r ) , a 1 = 0 , a p j = 0 ( 1 ≤ j < n ) . (34) The corresponding second component is
h = δ ( f ) , (35) h=\delta(f), \tag*{(35)} h = δ ( f ) , (35) and the action (33) is
σ ∗ b = h − 1 δ ( h ) + h σ ( b ) . (36) \sigma*b=h^{-1}\delta(h)+h\sigma(b). \tag*{(36)} σ ∗ b = h − 1 δ ( h ) + hσ ( b ) . (36)
Proof. Every automorphism σ \sigma σ of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) preserves its maximal ideal, so f = σ ( x ) f=\sigma(x) f = σ ( x ) has zero constant coefficient. The condition h δ σ = δ h\delta^\sigma=\delta h δ σ = δ is equivalent to
δ σ = h σ δ . (37) \delta\sigma=h\sigma\delta. \tag*{(37)} δ σ = hσ δ . (37)
Applying it to x x x gives h = δ ( f ) h=\delta(f) h = δ ( f ) . We prove σ ( x ( r ) ) = f ( r ) \sigma(x^{(r)})=f^{(r)} σ ( x ( r ) ) = f ( r ) by induction on r r r . The case r = 0 r=0 r = 0 is immediate. For 1 ≤ r < p n 1\leq r<p^n 1 ≤ r < p n , the induction hypothesis, (37) , and Lemma 2.2 give
δ ( σ ( x ( r ) ) − f ( r ) ) = h σ ( x ( r − 1 ) ) − δ ( f ) f ( r − 1 ) = 0. \delta\left(\sigma(x^{(r)})-f^{(r)}\right)=h\sigma(x^{(r-1)})-\delta(f)f^{(r-1)}=0. δ ( σ ( x ( r ) ) − f ( r ) ) = hσ ( x ( r − 1 ) ) − δ ( f ) f ( r − 1 ) = 0.
Since ker δ = k 1 \ker\delta=k1 ker δ = k 1 in O ( 1 ) \mathcal{O}(1) O ( 1 ) and both terms have zero constant coefficient, their difference is zero. Thus σ \sigma σ is admissible. Conversely, let σ \sigma σ be an admissible automorphism and write f = σ ( x ) = ∑ m = 1 p n − 1 a m x ( m ) f=\sigma(x)=\sum_{m=1}^{p^n-1}a_mx^{(m)} f = σ ( x ) = ∑ m = 1 p n − 1 a m x ( m ) . The divided-power expansion shows that f ( r ) f^{(r)} f ( r ) has no terms of degree less than r r r , and its degree-r r r coefficient is a 1 r a_1^r a 1 r . Thus the matrix of σ \sigma σ is triangular with diagonal 1 , a 1 , … , a 1 p n − 1 1,a_1,\ldots,a_1^{p^n-1} 1 , a 1 , … , a 1 p n − 1 . Invertibility gives a 1 ≠ 0 a_1\ne0 a 1 = 0 , so h = δ ( f ) h=\delta(f) h = δ ( f ) is a unit. The chain rule gives (4.6, and hence ( σ , h ) ∈ G n (\sigma,h)\in G_n ( σ , h ) ∈ G n . This proves the asserted identification of groups and (35) . Substituting δ σ = h − 1 δ \delta^\sigma=h^{-1}\delta δ σ = h − 1 δ in (33) gives (36) .
The restrictions on the coefficients of f f f arise from requiring f ( r ) ∈ O ( 1 ; n ) f^{(r)}\in\mathcal{O}(1;n) f ( r ) ∈ O ( 1 ; n ) for 0 ≤ r < p n 0\leq r<p^n 0 ≤ r < p n . We obtain the necessary restrictions by examining the coefficient of x ( p n ) x^{(p^n)} x ( p n ) . For any two elements u , v u,v u , v of O ( 1 ) \mathcal{O}(1) O ( 1 ) with zero constant coefficient, [ x ( p n ) ] ( u v ) = 0 [x^{(p^n)}](uv)=0 [ x ( p n ) ] ( uv ) = 0 , since ( p n i ) = 0 \binom{p^n}{i}=0 ( i p n ) = 0 in k k k for 0 < i < p n 0<i<p^n 0 < i < p n . Fix 1 ≤ j < n 1\leq j<n 1 ≤ j < n and put r = p n − j r=p^{n-j} r = p n − j and m = p j m=p^j m = p j , so that r m = p n rm=p^n r m = p n . Every mixed term in the divided-power expansion of f ( r ) f^{(r)} f ( r ) contains at least two factors with zero constant coefficient and hence contributes nothing to x ( p n ) x^{(p^n)} x ( p n ) . Among the pure terms, only ( a m x ( m ) ) ( r ) (a_mx^{(m)})^{(r)} ( a m x ( m ) ) ( r ) can have degree p n p^n p n . Admissibility therefore gives
0 = [ x ( p n ) ] f ( r ) = C j ‾ a p j p n − j , C j = ( p n ) ! r ! ( m ! ) r ∈ Z . (38) 0=[x^{(p^n)}]f^{(r)}=\overline{C_j}a_{p^j}^{p^{n-j}},\qquad C_j=\frac{(p^n)!}{r!(m!)^r}\in\mathbb{Z}. \tag*{(38)} 0 = [ x ( p n ) ] f ( r ) = C j a p j p n − j , C j = r ! ( m ! ) r ( p n )! ∈ Z . (38)
where C j ‾ \overline{C_j} C j is the image of C j C_j C j in k \mathsf{k} k . Counting multiples of p p p , p 2 , … p^2,\ldots p 2 , … shows that the exponents of p p p in the numerator and denominator of C j C_j C j are, respectively,
p n − 1 p − 1 and r − 1 p − 1 + r m − 1 p − 1 = r m − 1 p − 1 = p n − 1 p − 1 . \frac{p^n-1}{p-1}\qquad\text{and}\qquad\frac{r-1}{p-1}+r\frac{m-1}{p-1}=\frac{rm-1}{p-1}=\frac{p^n-1}{p-1}. p − 1 p n − 1 and p − 1 r − 1 + r p − 1 m − 1 = p − 1 r m − 1 = p − 1 p n − 1 .
Hence p ∤ C j p\nmid C_j p ∤ C j , so C j ‾ ≠ 0 \overline{C_j}\ne0 C j = 0 , and (38) implies a p j = 0 a_{p^j}=0 a p j = 0 .
Conversely, suppose that f f f satisfies (34) . To construct the corresponding automorphism, we first verify that f ( r ) ∈ O ( 1 ; n ) f^{(r)}\in\mathcal{O}(1;n) f ( r ) ∈ O ( 1 ; n ) for 0 ≤ r < p n 0\le r<p^n 0 ≤ r < p n . If 1 < m < p n 1<m<p^n 1 < m < p n is not a power of p p p , choose a nonzero digit m j m_j m j in its base-p p p expansion. Then m − p j > 0 m-p^j>0 m − p j > 0 , and Lucas’s congruence gives
x ( m ) = m j − 1 x ( p j ) x ( m − p j ) . x^{(m)}=m_j^{-1}x^{(p^j)}x^{(m-p^j)}. x ( m ) = m j − 1 x ( p j ) x ( m − p j ) .
Since ( x ( p j ) ) p = p ! ( x ( p j ) ) ( p ) = 0 (x^{(p^j)})^p=p!(x^{(p^j)})^{(p)}=0 ( x ( p j ) ) p = p ! ( x ( p j ) ) ( p ) = 0 , the identity ( u v ) ( s ) = u s v ( s ) (uv)^{(s)}=u^sv^{(s)} ( uv ) ( s ) = u s v ( s ) gives
( x ( m ) ) ( s ) = 0 ( s ≥ p ) , ( x ( m ) ) ( s ) = ( x ( m ) ) s s ! ∈ O ( 1 ; n ) ( 0 ≤ s < p ) . \left(x^{(m)}\right)^{(s)}=0\quad(s\ge p),\qquad
\left(x^{(m)}\right)^{(s)}=\frac{\left(x^{(m)}\right)^s}{s!}\in\mathcal{O}(1;n)\quad(0\le s<p). ( x ( m ) ) ( s ) = 0 ( s ≥ p ) , ( x ( m ) ) ( s ) = s ! ( x ( m ) ) s ∈ O ( 1 ; n ) ( 0 ≤ s < p ) .
In the divided-power expansion of f ( r ) f^{(r)} f ( r ) , every factor arising from a nonzero term a m x ( m ) a_mx^{(m)} a m x ( m ) with m > 1 m>1 m > 1 therefore lies in O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) . The factors arising from a 1 x a_1x a 1 x are a 1 s x ( s ) a_1^sx^{(s)} a 1 s x ( s ) with s ≤ r < p n s\le r<p^n s ≤ r < p n , so they also lie in O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) . Since O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) is a subalgebra, the claim follows.
We can therefore define σ ( x ( r ) ) = f ( r ) \sigma(x^{(r)})=f^{(r)} σ ( x ( r ) ) = f ( r ) for 0 ≤ r < p n 0\le r<p^n 0 ≤ r < p n . The divided-power product identity, together with ( r + s r ) = 0 \binom{r+s}{r}=0 ( r r + s ) = 0 when r , s < p n ≤ r + s r,s<p^n\le r+s r , s < p n ≤ r + s , shows that σ \sigma σ is an algebra endomorphism. The triangularity established above and a 1 ≠ 0 a_1\ne0 a 1 = 0 imply that σ \sigma σ is invertible. It is admissible by construction and uniquely determined by f f f . □
We write σ \sigma σ for ( σ , δ ( σ ( x ) ) ) ∈ G n (\sigma,\delta(\sigma(x)))\in G_n ( σ , δ ( σ ( x ))) ∈ G n and use the filtration F s F^s F s from (12) .
Lemma 4.2. Let σ ≠ Id \sigma\ne\operatorname{Id} σ = Id be an admissible automorphism , with coefficients a s a_s a s as in (34) and a 1 = 1 a_1=1 a 1 = 1 . Put r = min { s : 2 ≤ s < p n , a s ≠ 0 } r=\min\{s:2\le s<p^n,\ a_s\ne0\} r = min { s : 2 ≤ s < p n , a s = 0 } . Then, for every b ∈ O ( 1 ; n ) b\in\mathcal{O}(1;n) b ∈ O ( 1 ; n ) ,
σ ∗ b ≡ b + a r x ( r − 2 ) ( m o d F r − 1 ) . (39) \sigma*b\equiv b+a_rx^{(r-2)}\pmod{F^{r-1}}. \tag*{(39)} σ ∗ b ≡ b + a r x ( r − 2 ) ( mod F r − 1 ) . (39)
Proof. Put f = σ ( x ) f=\sigma(x) f = σ ( x ) and h = δ ( f ) h=\delta(f) h = δ ( f ) . We estimate the two terms in (36) separately. By the choice of r r r ,
h ≡ 1 + a r x ( r − 1 ) ( m o d F r ) , δ ( h ) ≡ a r x ( r − 2 ) ( m o d F r − 1 ) . h\equiv1+a_rx^{(r-1)}\pmod{F^r},\qquad
\delta(h)\equiv a_rx^{(r-2)}\pmod{F^{r-1}}. h ≡ 1 + a r x ( r − 1 ) ( mod F r ) , δ ( h ) ≡ a r x ( r − 2 ) ( mod F r − 1 ) .
Since h − 1 − 1 ∈ F r − 1 h^{-1}-1\in F^{r-1} h − 1 − 1 ∈ F r − 1 and r ≥ 2 r\ge2 r ≥ 2 ,
( h − 1 − 1 ) δ ( h ) ∈ F 2 r − 3 ⊆ F r − 1 , (h^{-1}-1)\delta(h)\in F^{2r-3}\subseteq F^{r-1}, ( h − 1 − 1 ) δ ( h ) ∈ F 2 r − 3 ⊆ F r − 1 ,
so
h − 1 δ ( h ) ≡ a r x ( r − 2 ) ( m o d F r − 1 ) . (40) h^{-1}\delta(h)\equiv a_rx^{(r-2)}\pmod{F^{r-1}}. \tag*{(40)} h − 1 δ ( h ) ≡ a r x ( r − 2 ) ( mod F r − 1 ) . (40)
For the term h σ ( b ) h\sigma(b) hσ ( b ) , write u = f − x ∈ F r u=f-x\in F^r u = f − x ∈ F r . The divided-power expansion in O ( 1 ) \mathcal{O}(1) O ( 1 ) gives, for 1 ≤ s < p n 1\le s<p^n 1 ≤ s < p n ,
f ( s ) − x ( s ) = ∑ i = 1 s x ( s − i ) u ( i ) . f^{(s)}-x^{(s)}=\sum_{i=1}^{s}x^{(s-i)}u^{(i)}. f ( s ) − x ( s ) = i = 1 ∑ s x ( s − i ) u ( i ) .
Every term in the i i i th summand has degree at least ( s − i ) + i r ≥ s + r − 1 (s-i)+ir\ge s+r-1 ( s − i ) + i r ≥ s + r − 1 ; hence σ ( x ( s ) ) − x ( s ) ∈ F s + r − 1 \sigma(x^{(s)})-x^{(s)}\in F^{s+r-1} σ ( x ( s ) ) − x ( s ) ∈ F s + r − 1 . Since σ ( 1 ) = 1 \sigma(1)=1 σ ( 1 ) = 1 , it follows that σ ( b ) − b ∈ F r \sigma(b)-b\in F^r σ ( b ) − b ∈ F r and h σ ( b ) − b ∈ F r − 1 h\sigma(b)-b\in F^{r-1} hσ ( b ) − b ∈ F r − 1 . Combining these inclusions with (40) proves (39) . □
Define the ( n + 1 ) (n+1) ( n + 1 ) -dimensional subspace
S n = ⨁ j = 1 n k x ( p j − 2 ) ⊕ k x ( p n − 1 ) . (41) S_n=\bigoplus_{j=1}^{n}\mathsf{k}x^{(p^j-2)}\oplus\mathsf{k}x^{(p^n-1)}. \tag*{(41)} S n = j = 1 ⨁ n k x ( p j − 2 ) ⊕ k x ( p n − 1 ) . (41)
and let G n 1 ⊆ G n G_n^1\subseteq G_n G n 1 ⊆ G n consist of the elements with a 1 = 1 a_1=1 a 1 = 1 in (34) . This is a subgroup because the coefficient a 1 a_1 a 1 is multiplicative under composition.
Proposition 4.3. For every b ∈ O ( 1 ; n ) b \in\mathcal{O}(1;n) b ∈ O ( 1 ; n ) , there is a unique g ∈ G n 1 g \in G_n^1 g ∈ G n 1 such that g ∗ b ∈ S n g*b \in S_n g ∗ b ∈ S n . For α , α ′ ∈ k n \alpha,\alpha' \in\mathbb{k}^n α , α ′ ∈ k n and β , β ′ ∈ k \beta,\beta' \in\mathbb{k} β , β ′ ∈ k , the elements b α , β b_{\alpha,\beta} b α , β and b α ′ , β ′ b_{\alpha',\beta'} b α ′ , β ′ in (1) lie in the same G n G_n G n -orbit if and only if there is c ∈ k × c \in\mathbb{k}^{\times} c ∈ k × such that
α j ′ = c p j − 1 α j ( 1 ≤ j ≤ n ) , β ′ = c p n β . (42) \alpha'_j=c^{p^j-1}\alpha_j \quad(1\leq j\leq n),\qquad\beta'=c^{p^n}\beta. \tag*{(42)} α j ′ = c p j − 1 α j ( 1 ≤ j ≤ n ) , β ′ = c p n β . (42) Every element of G n G_n G n carrying one point of S n S_n S n to another is ( ρ c , c ) (\rho_c,c) ( ρ c , c ) for some c ∈ k × c\in\mathbb{k}^{\times} c ∈ k × , where
ρ c ( x ( r ) ) = c r x ( r ) . \rho_c(x^{(r)})=c^r x^{(r)}. ρ c ( x ( r ) ) = c r x ( r ) .
Proof. We eliminate the coefficients outside the degrees occurring in S n S_n S n , starting with the lowest degree. For each integer m m m with 0 ≤ m ≤ p n − 3 0\leq m\leq p^n-3 0 ≤ m ≤ p n − 3 and
m ∉ { p j − 2 ∣ 1 ≤ j < n } , m\notin\{p^j-2\mid1\leq j<n\}, m ∈ / { p j − 2 ∣ 1 ≤ j < n } ,
Proposition 4.1 gives an admissible automorphism σ m \sigma_m σ m with
σ m ( x ) = x + c m x ( m + 2 ) . \sigma_m(x)=x+c_mx^{(m+2)}. σ m ( x ) = x + c m x ( m + 2 ) .
Lemma 4.2 gives
[ x ( s ) ] ( σ m ∗ b ) = { [ x ( s ) ] b , 0 ≤ s < m , [ x ( m ) ] b + c m , s = m . [x^{(s)}](\sigma_m*b)=
\begin{cases}
[x^{(s)}]b, & 0\leq s<m,\\
[x^{(m)}]b+c_m, & s=m.
\end{cases} [ x ( s ) ] ( σ m ∗ b ) = { [ x ( s ) ] b , [ x ( m ) ] b + c m , 0 ≤ s < m , s = m .
At each step, choose c m = − [ x ( m ) ] b c_m=-[x^{(m)}]b c m = − [ x ( m ) ] b for the current b b b . The displayed identity shows that this cancels the coefficient of degree m m m without changing the coefficients already eliminated. Take σ m = Id \sigma_m=\operatorname{Id} σ m = Id when c m = 0 c_m=0 c m = 0 . Their composition gives g ∈ G n 1 g\in G_n^1 g ∈ G n 1 with g ∗ b ∈ S n g*b\in S_n g ∗ b ∈ S n .
For uniqueness, we show that an element of G n 1 G_n^1 G n 1 carrying one point of S n S_n S n to another must be the identity. Suppose that b , b ′ ∈ S n b,b'\in S_n b , b ′ ∈ S n and that b ′ = σ ∗ b b'=\sigma*b b ′ = σ ∗ b , where σ ∈ G n 1 \sigma\in G_n^1 σ ∈ G n 1 . If σ \sigma σ is not the identity, put r = min { s ∣ 2 ≤ s < p n , a s ≠ 0 } r=\min\{s\mid2\leq s<p^n,\ a_s\ne0\} r = min { s ∣ 2 ≤ s < p n , a s = 0 } . By (34) , r ≠ p j r\ne p^j r = p j for 1 ≤ j < n 1\leq j<n 1 ≤ j < n . Since r ≤ p n − 1 r\leq p^n-1 r ≤ p n − 1 , we also have r − 2 ≤ p n − 3 r-2\leq p^n-3 r − 2 ≤ p n − 3 . Thus r − 2 r-2 r − 2 is not one of the degrees occurring in S n S_n S n . But Lemma 4.2 shows that the coefficient of x ( r − 2 ) x^{(r-2)} x ( r − 2 ) changes from 0 0 0 to a r ≠ 0 a_r\ne0 a r = 0 , a contradiction. Thus σ \sigma σ is the identity. If g 1 , g 2 ∈ G n 1 g_1,g_2\in G_n^1 g 1 , g 2 ∈ G n 1 both carry a given b ∈ O ( 1 ; n ) b\in\mathcal{O}(1;n) b ∈ O ( 1 ; n ) into S n S_n S n , then g 2 g 1 − 1 g_2g_1^{-1} g 2 g 1 − 1 sends g 1 ∗ b g_1*b g 1 ∗ b to g 2 ∗ b g_2*b g 2 ∗ b . It is therefore the identity, and g 1 = g 2 g_1=g_2 g 1 = g 2 .
It remains to account for automorphisms whose linear coefficient need not be 1 1 1 . For c ∈ k × c\in\mathbb{k}^{\times} c ∈ k × , let ρ c \rho_c ρ c be the admissible automorphism ρ c ( x ) = c x \rho_c(x)=cx ρ c ( x ) = c x . Here h = c h=c h = c , so ρ c ∗ b = c ρ c ( b ) \rho_c*b=c\rho_c(b) ρ c ∗ b = c ρ c ( b ) and
ρ c ∗ x ( s ) = c s + 1 x ( s ) . (43) \rho_c*x^{(s)}=c^{s+1}x^{(s)}. \tag*{(43)} ρ c ∗ x ( s ) = c s + 1 x ( s ) . (43)
In particular, ρ c \rho_c ρ c preserves S n S_n S n . Every admissible automorphism σ \sigma σ has a unique factorization σ = ρ c τ \sigma=\rho_c\tau σ = ρ c τ , with c = a 1 c=a_1 c = a 1 and τ = ρ c − 1 σ \tau=\rho_c^{-1}\sigma τ = ρ c − 1 σ ; the linear coefficient of τ ( x ) \tau(x) τ ( x ) is 1 1 1 . If b , b ′ ∈ S n b,b'\in S_n b , b ′ ∈ S n and b ′ = ( ρ c τ ) ∗ b b'=(\rho_c\tau)*b b ′ = ( ρ c τ ) ∗ b , then
τ ∗ b = ρ c − 1 ∗ b ′ ∈ S n ⟹ τ = Id . \tau*b=\rho_c^{-1}*b'\in S_n\quad\Longrightarrow\quad\tau=\operatorname{Id}. τ ∗ b = ρ c − 1 ∗ b ′ ∈ S n ⟹ τ = Id .
Thus every element carrying one point of S n S_n S n to another is a scaling, and (43) gives (42) . Conversely, ρ c \rho_c ρ c realizes every relation (42) . □
Theorem 4.4 (Theorme A). Every finite-dimensional simple Novikov algebra over k \mathbb{k} k is isomorphic to N ( α , β ) N(\alpha,\beta) N ( α , β ) for some n ≥ 0 n\geq0 n ≥ 0 , α ∈ k n \alpha\in\mathbb{k}^n α ∈ k n , and β ∈ k \beta\in\mathbb{k} β ∈ k , with β = 1 \beta=1 β = 1 if n = 0 n=0 n = 0 . Two such normal forms are isomorphic if and only if they have the same dimension and their parameters satisfy (42) for some c ∈ k × c\in\mathbb{k}^{\times} c ∈ k × . Every isomorphism N ( α , β ) → N ( α ′ , β ′ ) N(\alpha,\beta)\to N(\alpha',\beta') N ( α , β ) → N ( α ′ , β ′ ) is
ϕ c = c − 1 ρ c , ϕ c ( y i ) = c i y i ( − 1 ≤ i ≤ p n − 2 ) , \phi_c=c^{-1}\rho_c,\qquad\phi_c(y_i)=c^i y_i\quad(-1\leq i\leq p^n-2), ϕ c = c − 1 ρ c , ϕ c ( y i ) = c i y i ( − 1 ≤ i ≤ p n − 2 ) , for a unique such scalar c c c , and every such scalar gives an isomorphism. For n = 0 n=0 n = 0 , set ρ c = Id k \rho_c=\operatorname{Id}_{\mathbb{k}} ρ c = Id k .
Proof. For n = 0 n=0 n = 0 , the model is k \mathbb{k} k and β = β ′ = 1 \beta=\beta'=1 β = β ′ = 1 , so (42) gives c = 1 c=1 c = 1 . Suppose n > 0 n>0 n > 0 . By Proposition 3.3, every simple Novikov algebra of dimension p n p^n p n is isomorphic to O ( 1 ; n ) ( δ , b ) \mathcal{O}(1;n)(\delta,b) O ( 1 ; n ) ( δ , b ) . By Proposition 4.3, we may reduce b b b to b α , β b_{\alpha,\beta} b α , β . The divided-power product gives
x ( i + 1 ) δ ( x ( j + 1 ) ) = ( i + j + 1 j ) x ( i + j + 1 ) , x^{(i+1)}\delta\left(x^{(j+1)}\right)=\binom{i+j+1}{j}x^{(i+j+1)}, x ( i + 1 ) δ ( x ( j + 1 ) ) = ( j i + j + 1 ) x ( i + j + 1 ) ,
x ( i + 1 ) x ( j + 1 ) x ( p s − 2 ) = ( i + j + 2 i + 1 ) ( i + j + p s p s − 2 ) x ( i + j + p s ) , x^{(i+1)}x^{(j+1)}x^{(p^{s}-2)}=\binom{i+j+2}{i+1}\binom{i+j+p^{s}}{p^{s}-2}x^{(i+j+p^{s})}, x ( i + 1 ) x ( j + 1 ) x ( p s − 2 ) = ( i + 1 i + j + 2 ) ( p s − 2 i + j + p s ) x ( i + j + p s ) ,
x ( i + 1 ) x ( j + 1 ) x ( p n − 1 ) = δ i , − 1 δ j , − 1 x ( p n − 1 ) . x^{(i+1)}x^{(j+1)}x^{(p^{n}-1)}=\delta_{i,-1}\delta_{j,-1}x^{(p^{n}-1)}. x ( i + 1 ) x ( j + 1 ) x ( p n − 1 ) = δ i , − 1 δ j , − 1 x ( p n − 1 ) .
For j = − 1 j=-1 j = − 1 , both sides of the first equality are zero by our convention for binomial coefficients. In the last equality, every term with i > − 1 i>-1 i > − 1 or j > − 1 j>-1 j > − 1 has degree at least p n p^{n} p n and vanishes. With y i = x ( i + 1 ) y_i=x^{(i+1)} y i = x ( i + 1 ) , substituting these identities into u ∘ v = u ( δ ( v ) + b α , β v ) u\circ v=u(\delta(v)+b_{\alpha,\beta}v) u ∘ v = u ( δ ( v ) + b α , β v ) gives (2) .
By Theorem 2.8 and Proposition 4.3, the isomorphism criterion is (42) , and every isomorphism between normal forms is c − 1 ρ c c^{-1}\rho_c c − 1 ρ c . The dimension determines n n n , and ϕ c ( 1 ) = c − 1 1 \phi_c(1)=c^{-1}1 ϕ c ( 1 ) = c − 1 1 determines c c c . □
Further properties
Corollary 4.5. Let J = { j ∣ 1 ≤ j ≤ n , α j ≠ 0 } J=\{j\mid1\leq j\leq n,\ \alpha_j\ne0\} J = { j ∣ 1 ≤ j ≤ n , α j = 0 } . Via c ↦ c − 1 ρ c c\mapsto c^{-1}\rho_c c ↦ c − 1 ρ c , the k \mathbb{k} k -algebra automorphism group of N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) is
Aut ( N ( α , β ) ) ≅ { { 1 } , β ≠ 0 , k × , β = 0 , J = ∅ , { c ∈ k × : c p g − 1 = 1 } , β = 0 , J ≠ ∅ , g = gcd J . \operatorname{Aut}(\mathcal{N}(\alpha,\beta))\cong
\begin{cases}
\{1\}, & \beta\ne0,\\
\mathbb{k}^{\times}, & \beta=0,\ J=\varnothing,\\
\{c\in\mathbb{k}^{\times}:c^{p^{g}-1}=1\}, & \beta=0,\ J\ne\varnothing,\quad g=\gcd J.
\end{cases} Aut ( N ( α , β )) ≅ ⎩ ⎨ ⎧ { 1 } , k × , { c ∈ k × : c p g − 1 = 1 } , β = 0 , β = 0 , J = ∅ , β = 0 , J = ∅ , g = g cdJ . In the last case it is cyclic of order p g − 1 p^{g}-1 p g − 1 .
Proof. Theorem 4.4 identifies the automorphisms with scalars satisfying c p j − 1 = 1 c^{p^j-1}=1 c p j − 1 = 1 for j ∈ J j\in J j ∈ J and c p n β = β c^{p^n}\beta=\beta c p n β = β . If β ≠ 0 \beta\ne0 β = 0 , then ( c − 1 ) p n = 0 (c-1)^{p^n}=0 ( c − 1 ) p n = 0 , so c = 1 c=1 c = 1 . For β = 0 \beta=0 β = 0 and J ≠ ∅ J\ne\varnothing J = ∅ , use gcd j ∈ J ( p j − 1 ) = p gcd J − 1 \gcd_{j\in J}(p^j-1)=p^{\gcd J}-1 g cdj ∈ J ( p j − 1 ) = p g c d J − 1 . Finally, ϕ c ϕ d = ϕ c d \phi_c\phi_d=\phi_{cd} ϕ c ϕ d = ϕ c d , so this identification respects the group operation. □
Corollary 4.6. Every finite-dimensional simple Novikov algebra over k \mathbb{k} k contains a nonzero idempotent. In N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) , one may take 1 1 1 for n = 0 n=0 n = 0 and x = x ( 1 ) x=x^{(1)} x = x ( 1 ) for n > 0 n>0 n > 0 .
Proof. The case n = 0 n=0 n = 0 is immediate. For n > 0 n>0 n > 0 , one has δ ( x ) = 1 \delta(x)=1 δ ( x ) = 1 and x 2 = 2 x ( 2 ) x^2=2x^{(2)} x 2 = 2 x ( 2 ) . For 1 ≤ j ≤ n 1\leq j\leq n 1 ≤ j ≤ n , the divided-power product gives
x ( p j − 2 ) x ( 2 ) = ( p j 2 ) x ( p j ) = 0 , x ( p n − 1 ) x ( 2 ) = 0. x^{(p^j-2)}x^{(2)}=\binom{p^j}{2}x^{(p^j)}=0,\qquad x^{(p^n-1)}x^{(2)}=0. x ( p j − 2 ) x ( 2 ) = ( 2 p j ) x ( p j ) = 0 , x ( p n − 1 ) x ( 2 ) = 0. The first binomial coefficient vanishes because p > 2 p>2 p > 2 ; the second product vanishes by truncation. Thus b α , β x 2 = 0 b_{\alpha,\beta}x^2=0 b α , β x 2 = 0 , and x ∘ x = x δ ( x ) + b α , β x 2 = x x\circ x=x\delta(x)+b_{\alpha,\beta}x^2=x x ∘ x = x δ ( x ) + b α , β x 2 = x . □
Proposition 4.7. Let n ≥ 0 n\geq0 n ≥ 0 , α , α ′ ∈ k n \alpha,\alpha'\in\mathbb{k}^n α , α ′ ∈ k n , and β , β ′ ∈ k \beta,\beta'\in\mathbb{k} β , β ′ ∈ k , with β = β ′ = 1 \beta=\beta'=1 β = β ′ = 1 if n = 0 n=0 n = 0 .
(i) For the element
s α = 1 − ∑ j = 1 n α j x ( p j − 1 ) ∈ O ( 1 ; n ) , s_\alpha=1-\sum_{j=1}^{n}\alpha_jx^{(p^j-1)}\in\mathcal{O}(1;n), s α = 1 − j = 1 ∑ n α j x ( p j − 1 ) ∈ O ( 1 ; n ) , the characteristic and minimal polynomials of the left multiplication operator L s α L_{s_\alpha} L s α on N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) are both
P α , β ( X ) = X p n + ∑ j = 1 n α j p n − j X p n − j − β . (44) P_{\alpha,\beta}(X)=X^{p^n}+\sum_{j=1}^{n}\alpha_j^{p^{n-j}}X^{p^{n-j}}-\beta. \tag*{(44)} P α , β ( X ) = X p n + j = 1 ∑ n α j p n − j X p n − j − β . (44) (ii) The algebras N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) and N ( α ′ , β ′ ) \mathcal{N}(\alpha',\beta') N ( α ′ , β ′ ) are isomorphic if and only if there is c ∈ k × c\in\mathbb{k}^{\times} c ∈ k × such that
P α ′ , β ′ ( X ) = c p n P α , β ( X / c ) . (45) P_{\alpha',\beta'}(X)=c^{p^n}P_{\alpha,\beta}(X/c). \tag*{(45)} P α ′ , β ′ ( X ) = c p n P α , β ( X / c ) . (45)
Proof. (i) For n = 0 n=0 n = 0 , one has s α = 1 s_\alpha=1 s α = 1 , L s α = Id L_{s_\alpha}=\operatorname{Id} L s α = Id , and P α , 1 ( X ) = X − 1 P_{\alpha,1}(X)=X-1 P α , 1 ( X ) = X − 1 . Suppose n > 0 n>0 n > 0 , and put q = p n q=p^n q = p n . The lowest base-p p p digit of p i − 1 p^i-1 p i − 1 is p − 1 p-1 p − 1 , and that of p j − 2 p^j-2 p j − 2 is p − 2 p-2 p − 2 . Since p > 2 p>2 p > 2 , Lucas’s congruence gives
x ( p i − 1 ) x ( p j − 2 ) = 0 ( 1 ≤ i , j ≤ n ) . x^{(p^i-1)}x^{(p^j-2)}=0 \qquad(1\le i,j\le n). x ( p i − 1 ) x ( p j − 2 ) = 0 ( 1 ≤ i , j ≤ n ) .
Also x ( q − 1 ) x^{(q-1)} x ( q − 1 ) annihilates the maximal ideal. With b = b α , β b=b_{\alpha,\beta} b = b α , β and D = δ + m b D=\delta+m_b D = δ + m b , we therefore have
b s α = b , δ ( s α ) = − ∑ j = 1 n α j x ( p j − 2 ) , D ( s α ) = β x ( q − 1 ) . bs_\alpha=b,\qquad\delta(s_\alpha)=-\sum_{j=1}^{n}\alpha_jx^{(p^j-2)},\qquad D(s_\alpha)=\beta x^{(q-1)}. b s α = b , δ ( s α ) = − j = 1 ∑ n α j x ( p j − 2 ) , D ( s α ) = β x ( q − 1 ) .
Since s α ( 0 ) = 1 s_\alpha(0)=1 s α ( 0 ) = 1 , we may conjugate L s α L_{s_\alpha} L s α by multiplication with s α s_\alpha s α . The identities L s α = m s α D L_{s_\alpha}=m_{s_\alpha}D L s α = m s α D and D m s α = m s α δ + m D ( s α ) Dm_{s_\alpha}=m_{s_\alpha}\delta+m_{D(s_\alpha)} D m s α = m s α δ + m D ( s α ) give
m s α − 1 L s α m s α = m s α − 1 δ + β m x ( q − 1 ) = : E . (46) m_{s_\alpha}^{-1}L_{s_\alpha}m_{s_\alpha}=m_{s_\alpha}^{-1}\delta+\beta m_{x^{(q-1)}}=:E. \tag*{(46)} m s α − 1 L s α m s α = m s α − 1 δ + β m x ( q − 1 ) =: E . (46)
We use the ordered basis x ( 0 ) , … , x ( q − 1 ) x^{(0)},\ldots,x^{(q-1)} x ( 0 ) , … , x ( q − 1 ) , with rows and columns indexed by 0 , … , q − 1 0,\ldots,q-1 0 , … , q − 1 . For 1 ≤ r < q 1\le r<q 1 ≤ r < q and 1 ≤ j ≤ n 1\le j\le n 1 ≤ j ≤ n , Lucas’s congruence gives, in k \mathbb{k} k ,
( r + p j − 2 p j − 1 ) = { 1 , p j ∣ r − 1 , 0 , p j ∤ r − 1. \binom{r+p^j-2}{p^j-1}=
\begin{cases}
1,&p^j\mid r-1,\\
0,&p^j\nmid r-1.
\end{cases} ( p j − 1 r + p j − 2 ) = { 1 , 0 , p j ∣ r − 1 , p j ∤ r − 1.
Thus E ( 1 ) = β x ( q − 1 ) E(1)=\beta x^{(q-1)} E ( 1 ) = β x ( q − 1 ) and, for 1 ≤ r < q 1\le r<q 1 ≤ r < q ,
E ( x ( r ) ) = x ( r − 1 ) − ∑ 1 ≤ j ≤ n , p j ∣ r − 1 r + p j − 2 < q α j x ( r + p j − 2 ) . (47) E\left(x^{(r)}\right)=x^{(r-1)}-\sum_{\substack{1\le j\le n,\ p^j\mid r-1\\r+p^j-2<q}}\alpha_jx^{(r+p^j-2)}. \tag*{(47)} E ( x ( r ) ) = x ( r − 1 ) − 1 ≤ j ≤ n , p j ∣ r − 1 r + p j − 2 < q ∑ α j x ( r + p j − 2 ) . (47)
In particular, the entries immediately above the diagonal are 1 1 1 , and all entries farther above the diagonal vanish. To compare the minimal and characteristic polynomials, note that E ( x ( r ) ) ≡ x ( r − 1 ) ( m o d F r ) E(x^{(r)})\equiv x^{(r-1)}\pmod{F^r} E ( x ( r ) ) ≡ x ( r − 1 ) ( mod F r ) for 1 ≤ r < q 1\le r<q 1 ≤ r < q . Induction gives
E j ( x ( q − 1 ) ) ≡ x ( q − 1 − j ) ( m o d F q − j ) ( 0 ≤ j < q ) . E^j\left(x^{(q-1)}\right)\equiv x^{(q-1-j)}\pmod{F^{q-j}}\qquad(0\le j<q). E j ( x ( q − 1 ) ) ≡ x ( q − 1 − j ) ( mod F q − j ) ( 0 ≤ j < q ) .
These vectors form a basis, so the minimal polynomial of E E E has degree q q q and therefore equals its characteristic polynomial. The same holds for L s α L_{s_\alpha} L s α by (46) .
We now compute the characteristic polynomial recursively in n n n . Let C n ( X ) C_n(X) C n ( X ) be the characteristic polynomial of E E E when β = 0 \beta=0 β = 0 , and set C 0 ( X ) = X C_0(X)=X C 0 ( X ) = X . If also α n = 0 \alpha_n=0 α n = 0 , grouping consecutive basis vectors into p p p blocks of size p n − 1 p^{n-1} p n − 1 makes the matrix block upper triangular. A term indexed by j < n j<n j < n contributes an entry in column m p j + 1 mp^j+1 m p j + 1 and row ( m + 1 ) p j − 1 (m+1)p^j-1 ( m + 1 ) p j − 1 , so it stays within one such block. The only entries joining different blocks come from δ \delta δ and lie above the block diagonal. Translation by a multiple of p n − 1 p^{n-1} p n − 1 preserves the divisibility conditions in (47) , so each diagonal block is the matrix for ( α 1 , … , α n − 1 ) (\alpha_1,\ldots,\alpha_{n-1}) ( α 1 , … , α n − 1 ) in dimension p n − 1 p^{n-1} p n − 1 . The characteristic polynomial is therefore C n − 1 ( X ) p C_{n-1}(X)^p C n − 1 ( X ) p .
Restoring α n \alpha_n α n changes only the entry in row q − 1 q-1 q − 1 , column 1 1 1 by − α n -\alpha_n − α n , while restoring β \beta β changes only row q − 1 q-1 q − 1 , column 0 0 0 by β \beta β . Since both changes are in the last row, det ( X Id − E ) \det(X\operatorname{Id}-E) det ( X Id − E ) is affine linear in α n \alpha_n α n and β \beta β , with cofactors independent of both parameters. The minor for column 0 0 0 is lower triangular with diagonal entries − 1 -1 − 1 . For column 1 1 1 , expansion along column 0 0 0 gives a factor X X X and a lower triangular minor with diagonal entries − 1 -1 − 1 . Thus the cofactors are
cof q − 1 , 0 ( X Id − E ) = ( − 1 ) q − 1 ( − 1 ) q − 1 = 1 , cof q − 1 , 1 ( X Id − E ) = ( − 1 ) q X ( − 1 ) q − 2 = X . \operatorname{cof}_{q-1,0}(X\operatorname{Id}-E)=(-1)^{q-1}(-1)^{q-1}=1,\qquad\operatorname{cof}_{q-1,1}(X\operatorname{Id}-E)=(-1)^qX(-1)^{q-2}=X. cof q − 1 , 0 ( X Id − E ) = ( − 1 ) q − 1 ( − 1 ) q − 1 = 1 , cof q − 1 , 1 ( X Id − E ) = ( − 1 ) q X ( − 1 ) q − 2 = X .
Thus
C n ( X ) = C n − 1 ( X ) p + α n X = X p n + ∑ j = 1 n α j p n − j X p n − j , C_n(X)=C_{n-1}(X)^p+\alpha_nX=X^{p^n}+\sum_{j=1}^{n}\alpha_j^{p^{n-j}}X^{p^{n-j}}, C n ( X ) = C n − 1 ( X ) p + α n X = X p n + j = 1 ∑ n α j p n − j X p n − j ,
det ( X Id − E ) = C n ( X ) − β , \det(X\operatorname{Id}-E)=C_n(X)-\beta, det ( X Id − E ) = C n ( X ) − β ,
which proves (44) .
(ii) For two normal forms of dimension p n p^n p n , (45) is equivalent to
( α j ′ ) p n − j = c p n − p n − j α j p n − j , β ′ = c p n β . (\alpha'_j)^{p^{n-j}}=c^{p^n-p^{n-j}}\alpha_j^{p^{n-j}},\qquad\beta'=c^{p^n}\beta. ( α j ′ ) p n − j = c p n − p n − j α j p n − j , β ′ = c p n β .
Since the Frobenius map t ↦ t p t\mapsto t^p t ↦ t p on k \mathbb{k} k is injective, the first equalities give α j ′ = c p j − 1 α j \alpha'_j=c^{p^j-1}\alpha_j α j ′ = c p j − 1 α j . Theorem 4.4 now proves (ii). □
We can also identify its normal form by computing the characteristic polynomial of L 1 L_1 L 1 . On each p p p -dimensional invariant subspace ( 1 + x 2 ) j k [ x 1 ] / ( x 1 p ) (1+x_2)^j k[x_1]/(x_1^p) ( 1 + x 2 ) j k [ x 1 ] / ( x 1 p ) , 0 ≤ j < p 0\leq j<p 0 ≤ j < p , we have ∂ ∗ = − j Id \partial_*=-j\,\operatorname{Id} ∂ ∗ = − j Id and hence ( ∂ ∗ + j Id ) p = 0 (\partial_*+j\,\operatorname{Id})^p=0 ( ∂ ∗ + j Id ) p = 0 . Its characteristic polynomial is therefore ( X + j ) p = X p + j (X+j)^p=X^p+j ( X + j ) p = X p + j . Consequently,
det ( X Id − L 1 ) = ∏ j = 0 p − 1 ( X p + j ) = X p 2 − X p . (50) \det(X\,\operatorname{Id}-L_1)=\prod_{j=0}^{p-1}(X^p+j)=X^{p^2}-X^p. \tag*{(50)} det ( X Id − L 1 ) = j = 0 ∏ p − 1 ( X p + j ) = X p 2 − X p . (50)
In its normal form, the right annihilator is ker D \ker D ker D , where D = δ + m b α , β D=\delta+m_{b_{\alpha,\beta}} D = δ + m b α , β . Let s α s_\alpha s α be as in Proposition 4.7. Since s α s_\alpha s α is a unit and P α , β ( 0 ) = − β P_{\alpha,\beta}(0)=-\beta P α , β ( 0 ) = − β , the identity L s α = m s α D L_{s_\alpha}=m_{s_\alpha}D L s α = m s α D shows that ker D ≠ 0 \ker D\ne0 ker D = 0 precisely when β = 0 \beta=0 β = 0 . In that case D ( s α u ) = s α δ ( u ) D(s_\alpha u)=s_\alpha\delta(u) D ( s α u ) = s α δ ( u ) for every u ∈ O ( 1 ; 2 ) u\in\mathcal{O}(1;2) u ∈ O ( 1 ; 2 ) , so ker D = s α ker δ = k s α \ker D=s_\alpha\ker\delta=ks_\alpha ker D = s α ker δ = k s α .
An isomorphism N ∗ → N ( α , 0 ) \mathcal{N}_*\to\mathcal{N}(\alpha,0) N ∗ → N ( α , 0 ) therefore sends 1 1 1 to c s α cs_\alpha c s α for some c ∈ k × c\in k^\times c ∈ k × . Comparing characteristic polynomials gives
X p 2 − X p = c p P α , 0 ( X / c ) , α 2 = 0 , c p − 1 α 1 = − 1. X^{p^2}-X^p=c^pP_{\alpha,0}(X/c),\qquad\alpha_2=0,\qquad c^{p-1}\alpha_1=-1. X p 2 − X p = c p P α , 0 ( X / c ) , α 2 = 0 , c p − 1 α 1 = − 1.
Theorem 4.4 therefore gives N ∗ ≅ N ( ( − 1 , 0 ) , 0 ) \mathcal{N}_*\cong\mathcal{N}((-1,0),0) N ∗ ≅ N (( − 1 , 0 ) , 0 ) , which lies outside the family (48) .
The same example locates the gap in Osborn’s reduction. Taking the right identity e = x 1 e=x_1 e = x 1 and u = 1 u=1 u = 1 in the construction of [14 Lemmas 3.5–3.6, pp. 2740–2742] gives the auxiliary algebra A 0 = K x 1 ≅ K \mathcal{A}_0=Kx_1\cong K A 0 = K x 1 ≅ K , with product ( c x 1 ) ∘ ( d x 1 ) = c d x 1 (cx_1)\circ(dx_1)=cdx_1 ( c x 1 ) ∘ ( d x 1 ) = c d x 1 for c , d ∈ K c,d\in K c , d ∈ K . On p. 2742, the auxiliary derivation ∂ \partial ∂ is defined by
u ∘ ( u ∘ t ) = v ∘ ∂ ( t ) ( t ∈ A 0 ) , v = x 1 p − 1 ( p − 1 ) ! = − x 1 p − 1 . u\circ(u\circ t)=v\circ\partial(t)\quad(t\in\mathcal{A}_0),\qquad v=\frac{x_1^{p-1}}{(p-1)!}=-x_1^{p-1}. u ∘ ( u ∘ t ) = v ∘ ∂ ( t ) ( t ∈ A 0 ) , v = ( p − 1 )! x 1 p − 1 = − x 1 p − 1 .
Here u , v u,v u , v are denoted by d , x p − 2 d,x_{p-2} d , x p − 2 in that proof. For c ∈ K c\in K c ∈ K , we have
u ∘ ( u ∘ ( c x 1 ) ) = x 1 p − 1 T ( c ) = v ∘ ( − T ( c ) x 1 ) . u\circ(u\circ(cx_1))=x_1^{p-1}T(c)=v\circ(-T(c)x_1). u ∘ ( u ∘ ( c x 1 )) = x 1 p − 1 T ( c ) = v ∘ ( − T ( c ) x 1 ) .
Since t ↦ v ∘ t t\mapsto v\circ t t ↦ v ∘ t is injective on A 0 \mathcal{A}_0 A 0 , this gives
∂ ( c x 1 ) = − T ( c ) x 1 ( c ∈ K ) , ∂ p = ∂ ≠ 0. (51) \partial(cx_1)=-T(c)x_1\quad(c\in K),\qquad\partial^p=\partial\ne0. \tag*{(51)} ∂ ( c x 1 ) = − T ( c ) x 1 ( c ∈ K ) , ∂ p = ∂ = 0. (51)
Since T T T is a unit multiple of ∂ 2 \partial_2 ∂ 2 , the algebra A 0 \mathcal{A}_0 A 0 is ∂ \partial ∂ -simple, and ker ∂ = k e \ker\partial=ke ker ∂ = k e . After proving ∂ \partial ∂ -simplicity, the argument uses a basis { t ( k ) : 0 ≤ k < p r } \{t^{(k)}:0\leq k<p^r\} { t ( k ) : 0 ≤ k < p r } satisfying
t ( 0 ) = e , ∂ ( t ( k ) ) = t ( k − 1 ) ( 1 ≤ k < p r ) . t^{(0)}=e,\qquad\partial(t^{(k)})=t^{(k-1)}\quad(1\leq k<p^r). t ( 0 ) = e , ∂ ( t ( k ) ) = t ( k − 1 ) ( 1 ≤ k < p r ) .
Here dim A 0 = p \dim\mathcal{A}_0=p dim A 0 = p , so r = 1 r=1 r = 1 and this would force ∂ p = 0 \partial^p=0 ∂ p = 0 , contrary to (51) . Thus ∂ \partial ∂ -simplicity does not justify this choice of basis. The use of [14 Lemma 3.6] immediately before [23 Theorem 2.9, p. 914] brings the same gap into the latter reduction.
Novikov–Poisson algebras with simple Novikov part
Definition 4.10 ([23 , 24 ] ). A Novikov–Poisson algebra is a vector space A \mathcal{A} A with a commutative associative product ⋅ \cdot ⋅ and a Novikov product ∘ \circ ∘ satisfying
( x ⋅ y ) ∘ z = x ⋅ ( y ∘ z ) , (x\cdot y)\circ z=x\cdot(y\circ z), ( x ⋅ y ) ∘ z = x ⋅ ( y ∘ z ) , ( x ∘ y ) ⋅ z − x ∘ ( y ⋅ z ) = ( y ∘ x ) ⋅ z − y ∘ ( x ⋅ z ) (x\circ y)\cdot z-x\circ(y\cdot z)=(y\circ x)\cdot z-y\circ(x\cdot z) ( x ∘ y ) ⋅ z − x ∘ ( y ⋅ z ) = ( y ∘ x ) ⋅ z − y ∘ ( x ⋅ z ) for all x , y , z ∈ A x,y,z\in\mathcal{A} x , y , z ∈ A .
Corollary 4.11. Every finite-dimensional Novikov–Poisson algebra whose Novikov part is simple is isomorphic to O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) with products
f ∘ g = f ( δ ( g ) + b α , β g ) , f ⋅ g = u γ f g , u γ = ∑ r = 0 p n − 1 γ r x ( r ) , (52) f\circ g=f\bigl(\delta(g)+b_{\alpha,\beta}g\bigr),\qquad f\cdot g=u_\gamma fg,\qquad u_\gamma=\sum_{r=0}^{p^n-1}\gamma_r x^{(r)}, \tag*{(52)} f ∘ g = f ( δ ( g ) + b α , β g ) , f ⋅ g = u γ f g , u γ = r = 0 ∑ p n − 1 γ r x ( r ) , (52) where ( n , α , β ) (n,\alpha,\beta) ( n , α , β ) are as in Theorem 4.4, γ ∈ k p n \gamma\in k^{p^n} γ ∈ k p n is arbitrary, and juxtaposition denotes the divided-power product. Two such algebras are isomorphic if and only if they have the same n n n and there exists c ∈ k × c\in k^\times c ∈ k × such that
α j ′ = c p j − 1 α j , β ′ = c p n β , γ r ′ = c r + 1 γ r . (53) \alpha'_j=c^{p^j-1}\alpha_j,\qquad\beta'=c^{p^n}\beta,\qquad\gamma'_r=c^{r+1}\gamma_r. \tag*{(53)} α j ′ = c p j − 1 α j , β ′ = c p n β , γ r ′ = c r + 1 γ r . (53)
for 1 ≤ j ≤ n 1 \le j \le n 1 ≤ j ≤ n and 0 ≤ r < p n 0 \le r < p^n 0 ≤ r < p n .
Proof. By Theorem 4.4 , we may identify the Novikov part with N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) . Put M f ( g ) = f ⋅ g M_f(g)=f\cdot g M f ( g ) = f ⋅ g . The first compatibility identity gives [ M f , R g ] = 0 [M_f,R_g]=0 [ M f , R g ] = 0 . By Lemma 2.7 , the right multiplications generate all multiplication operators of O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n ) . Hence each M f M_f M f commutes with these operators, and commutativity gives
f ⋅ g = g M f ( 1 ) = g M 1 ( f ) = f g M 1 ( 1 ) . f\cdot g=gM_f(1)=gM_1(f)=fgM_1(1). f ⋅ g = g M f ( 1 ) = g M 1 ( f ) = f g M 1 ( 1 ) .
Thus f ⋅ g = u f g f\cdot g=ufg f ⋅ g = u f g for a unique u ∈ O ( 1 ; n ) u\in\mathcal{O}(1;n) u ∈ O ( 1 ; n ) . Conversely, every such product is commutative and associative and satisfies the first compatibility identity. The second follows from
( f ∘ g ) ⋅ h − f ∘ ( g ⋅ h ) = − f g δ ( u h ) . (f\circ g)\cdot h-f\circ(g\cdot h)=-fg\,\delta(uh). ( f ∘ g ) ⋅ h − f ∘ ( g ⋅ h ) = − f g δ ( u h ) .
Finally, by Theorem 4.4 , all isomorphisms between the Novikov parts are φ c = c − 1 ρ c \varphi_c=c^{-1}\rho_c φ c = c − 1 ρ c , where ρ c ( x ( r ) ) = c r x ( r ) \rho_c(x^{(r)})=c^r x^{(r)} ρ c ( x ( r ) ) = c r x ( r ) . Preservation of ⋅ \cdot ⋅ is equivalent to u γ = c ρ c ( u γ ) u_\gamma=c\rho_c(u_\gamma) u γ = c ρ c ( u γ ) , which gives (53) . □
Classification of irreducible bimodules
For the algebras in Theorem A, we recover an action of the underlying associative algebra on every nontrivial irreducible bimodule and prove that each such bimodule is finite-dimensional. We then classify these bimodules and determine their isomorphisms.
Definition 5.1. Let N \mathcal{N} N be a Novikov algebra.
(i) A bimodule over N \mathcal{N} N is a vector space V V V with linear maps ℓ , r : N → End k ( V ) \ell,r:\mathcal{N}\to\operatorname{End}_k(V) ℓ , r : N → End k ( V ) such that the split null extension N ⊕ V \mathcal{N}\oplus V N ⊕ V , with product
( x + u ) ∘ ( y + v ) = x ∘ y + ℓ ( x ) v + r ( y ) u , (x+u)\circ(y+v)=x\circ y+\ell(x)v+r(y)u, ( x + u ) ∘ ( y + v ) = x ∘ y + ℓ ( x ) v + r ( y ) u , is a Novikov algebra. Equivalently, for all x , y ∈ N x,y\in\mathcal{N} x , y ∈ N ,
ℓ ( x ∘ y ) = r ( y ) ℓ ( x ) , (54) \ell(x\circ y)=r(y)\ell(x), \tag*{(54)} ℓ ( x ∘ y ) = r ( y ) ℓ ( x ) , (54) [ r ( x ) , r ( y ) ] = 0 , (55) [r(x),r(y)]=0, \tag*{(55)} [ r ( x ) , r ( y )] = 0 , (55) [ ℓ ( x ) , ℓ ( y ) ] = ℓ ( x ∘ y − y ∘ x ) , (56) [\ell(x),\ell(y)]=\ell(x\circ y-y\circ x), \tag*{(56)} [ ℓ ( x ) , ℓ ( y )] = ℓ ( x ∘ y − y ∘ x ) , (56) [ ℓ ( x ) , r ( y ) ] = r ( x ∘ y ) − r ( y ) r ( x ) . (57) [\ell(x),r(y)]=r(x\circ y)-r(y)r(x). \tag*{(57)} [ ℓ ( x ) , r ( y )] = r ( x ∘ y ) − r ( y ) r ( x ) . (57) (ii) A nonzero bimodule is irreducible if it has no nonzero proper subspace stable under both actions.
(iii) A bimodule is nontrivial if its two actions are not both zero.
(iv) The regular bimodule is N \mathcal{N} N with ℓ ( x ) = L x \ell(x)=L_x ℓ ( x ) = L x and r ( x ) = R x r(x)=R_x r ( x ) = R x .
The only trivial irreducible bimodule, up to isomorphism, is the one-dimensional bimodule with both actions zero. We do not assume that bimodules are finite-dimensional.
Fix n ≥ 0 n\ge0 n ≥ 0 and write
A = O ( 1 ; n ) , D = δ + m b , N ( b ) = A ( δ , b ) ( b ∈ A ) . \mathcal{A}=\mathcal{O}(1;n),\qquad D=\delta+m_b,\qquad\mathcal{N}(b)=\mathcal{A}(\delta,b)\quad(b\in\mathcal{A}). A = O ( 1 ; n ) , D = δ + m b , N ( b ) = A ( δ , b ) ( b ∈ A ) .
When n = 0 n=0 n = 0 , take b = 1 b=1 b = 1 . For the normal forms in Theorem A, we have N ( α , β ) = N ( b α , β ) \mathcal{N}(\alpha,\beta)=\mathcal{N}(b_{\alpha,\beta}) N ( α , β ) = N ( b α , β ) . By Lemma 2.7 , N ( b ) \mathcal{N}(b) N ( b ) is simple and its right multiplications generate all multiplication operators on A \mathcal{A} A .
Lemma 5.2. Let V V V be a nontrivial irreducible N ( b ) \mathcal{N}(b) N ( b ) -bimodule and put T = ℓ ( 1 ) T=\ell(1) T = ℓ ( 1 ) . There is a unique injective unital algebra homomorphism π : A → End k ( V ) \pi:\mathcal{A}\to\operatorname{End}_k(V) π : A → End k ( V ) such that
ℓ ( x ) = π ( x ) T , r ( x ) = π ( D ( x ) ) , (58) \ell(x)=\pi(x)T,\qquad r(x)=\pi(D(x)), \tag*{(58)} ℓ ( x ) = π ( x ) T , r ( x ) = π ( D ( x )) , (58) [ T , π ( a ) ] = π ( δ ( a ) ) ( a ∈ A ) . (59) [T,\pi(a)]=\pi(\delta(a))\qquad(a\in\mathcal{A}). \tag*{(59)} [ T , π ( a )] = π ( δ ( a )) ( a ∈ A ) . (59) A subspace of V V V is a subbimodule if and only if it is stable under π ( A ) \pi(\mathcal{A}) π ( A ) and T T T .
Proof. For n = 0 n=0 n = 0 , put L = ℓ ( 1 ) L=\ell(1) L = ℓ ( 1 ) and R = r ( 1 ) R=r(1) R = r ( 1 ) . The bimodule identities give R L = L RL=L R L = L and [ L , R ] = R − R 2 [L,R]=R-R^2 [ L , R ] = R − R 2 . The subspace ker ( R − Id ) \ker(R-\operatorname{Id}) ker ( R − Id ) is stable under L L L and R R R and is nonzero: it contains im L \operatorname{im}L im L if L ≠ 0 L\ne0 L = 0 , and im R ≠ 0 \operatorname{im}R\ne0 im R = 0 if L = 0 L=0 L = 0 , since then R 2 = R R^2=R R 2 = R . Irreducibility gives R = Id R=\operatorname{Id} R = Id , so take π ( a ) = a Id \pi(a)=a\operatorname{Id} π ( a ) = a Id .
Suppose now that n > 0 n>0 n > 0 . To recover an associative action on V V V , we first show that the images of the left actions span V V V . Let U = ∑ x ∈ A ℓ ( x ) V U=\sum_{x\in A}\ell(x)V U = ∑ x ∈ A ℓ ( x ) V . The bimodule identities give
r ( y ) ℓ ( x ) = ℓ ( x ∘ y ) , ℓ ( y ) ℓ ( x ) = ℓ ( x ) ℓ ( y ) + ℓ ( y ∘ x − x ∘ y ) . r(y)\ell(x)=\ell(x\mathbin{\circ}y),\qquad\ell(y)\ell(x)=\ell(x)\ell(y)+\ell(y\mathbin{\circ}x-x\mathbin{\circ}y). r ( y ) ℓ ( x ) = ℓ ( x ∘ y ) , ℓ ( y ) ℓ ( x ) = ℓ ( x ) ℓ ( y ) + ℓ ( y ∘ x − x ∘ y ) .
so U U U is a subbimodule. If U = 0 U=0 U = 0 , then ℓ = 0 \ell=0 ℓ = 0 , and (55) and (57) make r r r a nonzero homomorphism into a commutative associative algebra. Simplicity makes r r r injective, contradicting 1 ∘ x ( 1 ) − x ( 1 ) ∘ 1 = 1 1\mathbin{\circ}x^{(1)}-x^{(1)}\mathbin{\circ}1=1 1 ∘ x ( 1 ) − x ( 1 ) ∘ 1 = 1 . Thus U = V U=V U = V .
We can now transfer polynomial relations from the regular right multiplications to the right actions on V V V . As in [15 ] and [25 proof of Lemma 4.1] , repeated use of r ( u ) ℓ ( x ) = ℓ ( R u x ) r(u)\ell(x)=\ell(R_u x) r ( u ) ℓ ( x ) = ℓ ( R u x ) gives, for u 1 , … , u m ∈ A u_1,\ldots,u_m\in A u 1 , … , u m ∈ A and P ∈ k [ t 1 , … , t m ] P\in k[t_1,\ldots,t_m] P ∈ k [ t 1 , … , t m ] ,
P ( r ( u 1 ) , … , r ( u m ) ) ℓ ( x ) = ℓ ( P ( R u 1 , … , R u m ) x ) . (60) P(r(u_1),\ldots,r(u_m))\ell(x)=\ell\bigl(P(R_{u_1},\ldots,R_{u_m})x\bigr). \tag*{(60)} P ( r ( u 1 ) , … , r ( u m )) ℓ ( x ) = ℓ ( P ( R u 1 , … , R u m ) x ) . (60)
If P ( R u 1 , … , R u m ) = 0 P(R_{u_1},\ldots,R_{u_m})=0 P ( R u 1 , … , R u m ) = 0 , then P ( r ( u 1 ) , … , r ( u m ) ) P(r(u_1),\ldots,r(u_m)) P ( r ( u 1 ) , … , r ( u m )) annihilates U = V U=V U = V . Since R u = m D ( u ) R_u=m_{D(u)} R u = m D ( u ) , we have
P ( R u 1 , … , R u m ) = m P ( D ( u 1 ) , … , D ( u m ) ) . P(R_{u_1},\ldots,R_{u_m})=m_{P(D(u_1),\ldots,D(u_m))}. P ( R u 1 , … , R u m ) = m P ( D ( u 1 ) , … , D ( u m )) .
It follows that every polynomial relation among the D ( u ) D(u) D ( u ) holds among the r ( u ) r(u) r ( u ) . Since D ( A ) D(A) D ( A ) generates A A A by Lemma 2.7, the assignment D ( y ) ↦ r ( y ) D(y)\mapsto r(y) D ( y ) ↦ r ( y ) extends uniquely to a unital algebra homomorphism π : A → End k ( V ) \pi:A\to\operatorname{End}_k(V) π : A → End k ( V ) . Explicitly,
π ( P ( D ( u 1 ) , … , D ( u m ) ) ) = P ( r ( u 1 ) , … , r ( u m ) ) , π ( 1 ) = Id V . \pi\bigl(P(D(u_1),\ldots,D(u_m))\bigr)=P(r(u_1),\ldots,r(u_m)),\qquad\pi(1)=\operatorname{Id}_V. π ( P ( D ( u 1 ) , … , D ( u m )) ) = P ( r ( u 1 ) , … , r ( u m )) , π ( 1 ) = Id V .
Equation (60) also gives π ( a ) ℓ ( x ) = ℓ ( a x ) \pi(a)\ell(x)=\ell(ax) π ( a ) ℓ ( x ) = ℓ ( a x ) . Taking x = 1 x=1 x = 1 yields ℓ ( a ) = π ( a ) T \ell(a)=\pi(a)T ℓ ( a ) = π ( a ) T .
The remaining commutator relation follows from (56) , which gives [ T , ℓ ( x ) ] = ℓ ( δ ( x ) ) [T,\ell(x)]=\ell(\delta(x)) [ T , ℓ ( x )] = ℓ ( δ ( x )) . Using π ( a ) ℓ ( x ) = ℓ ( a x ) \pi(a)\ell(x)=\ell(ax) π ( a ) ℓ ( x ) = ℓ ( a x ) , we obtain
[ T , π ( a ) ] ℓ ( x ) = ℓ ( δ ( a x ) ) − ℓ ( a δ ( x ) ) = π ( δ ( a ) ) ℓ ( x ) . [T,\pi(a)]\ell(x)=\ell(\delta(ax))-\ell(a\delta(x))=\pi(\delta(a))\ell(x). [ T , π ( a )] ℓ ( x ) = ℓ ( δ ( a x )) − ℓ ( a δ ( x )) = π ( δ ( a )) ℓ ( x ) .
Since U = V U=V U = V , this proves (59) . This relation also proves faithfulness: ker π \ker\pi ker π is a proper δ \delta δ -stable ideal and is zero by δ \delta δ -simplicity. The final assertion follows because π ( A ) \pi(A) π ( A ) is generated by the right actions and ℓ ( x ) = π ( x ) T \ell(x)=\pi(x)T ℓ ( x ) = π ( x ) T .
Lemma 5.3. Every nontrivial irreducible N ( b ) \mathcal{N}(b) N ( b ) -bimodule is finite-dimensional.
Proof. Let V V V be such a bimodule, with π \pi π and T = ℓ ( 1 ) T=\ell(1) T = ℓ ( 1 ) as in Lemma 5.2. Write a v = π ( a ) v av=\pi(a)v a v = π ( a ) v . Let m \mathfrak{m} m be the maximal ideal of A = O ( 1 ; n ) A=\mathcal{O}(1;n) A = O ( 1 ; n ) . We first use its nilpotence to choose a vector that will generate V V V under T T T . Take the largest s ≥ 0 s\ge0 s ≥ 0 such that m s V ≠ 0 \mathfrak{m}^sV\ne0 m s V = 0 and choose 0 ≠ v ∈ m s V 0\ne v\in\mathfrak{m}^sV 0 = v ∈ m s V . Then m v = 0 \mathfrak{m}v=0 m v = 0 , so a v = a ( 0 ) v av=a(0)v a v = a ( 0 ) v for every a ∈ A a\in A a ∈ A . Induction using a T = T a − δ ( a ) aT=Ta-\delta(a) a T = T a − δ ( a ) gives, for r ≥ 0 r\ge0 r ≥ 0 ,
a T r v = ∑ j = 0 r ( − 1 ) j ( r j ) T r − j δ j ( a ) v = ∑ j = 0 r ( − 1 ) j ( r j ) δ j ( a ) ( 0 ) T r − j v . aT^rv=\sum_{j=0}^{r}(-1)^j\binom{r}{j}T^{r-j}\delta^j(a)v
=\sum_{j=0}^{r}(-1)^j\binom{r}{j}\delta^j(a)(0)T^{r-j}v. a T r v = j = 0 ∑ r ( − 1 ) j ( j r ) T r − j δ j ( a ) v = j = 0 ∑ r ( − 1 ) j ( j r ) δ j ( a ) ( 0 ) T r − j v .
The subspace k [ T ] v k[T]v k [ T ] v is nonzero and stable under A A A and T T T , so it is a subbimodule and V = k [ T ] v V=k[T]v V = k [ T ] v by irreducibility.
Put q = p n q=p^n q = p n . If V V V were infinite-dimensional, the vectors T r v T^rv T r v would be linearly independent; any polynomial relation would express all sufficiently high powers in terms of finitely many lower ones. In that case
T q V = span k { T r v : r ≥ q } . T^qV=\operatorname{span}_k\{T^rv:r\ge q\}. T q V = span k { T r v : r ≥ q } .
would be nonzero and would not contain v v v . To contradict irreducibility, it remains to check stability under A \mathcal{A} A . Repeated use of (59) gives
T q π ( a ) = ∑ j = 0 q ( q j ) π ( δ j ( a ) ) T q − j = π ( a ) T q , T^{q}\pi(a)=\sum_{j=0}^{q}\binom{q}{j}\pi(\delta^{j}(a))T^{q-j}=\pi(a)T^{q}, T q π ( a ) = j = 0 ∑ q ( j q ) π ( δ j ( a )) T q − j = π ( a ) T q ,
since ( q j ) = 0 \binom{q}{j}=0 ( j q ) = 0 in k \mathbb{k} k for 0 < j < q 0<j<q 0 < j < q and δ q = 0 \delta^{q}=0 δ q = 0 . Hence T q V T^{q}V T q V is stable under A \mathcal{A} A and T T T . It is a nonzero proper subbimodule, contradicting irreducibility. □
For λ ∈ k \lambda\in\mathbb{k} λ ∈ k , let M ( b , λ ) \mathcal{M}(b,\lambda) M ( b , λ ) be the vector space A \mathcal{A} A with linear operators
ℓ λ ( x ) v = x δ ( v ) + λ x v , (61) \ell_{\lambda}(x)v=x\delta(v)+\lambda xv, \tag*{(61)} ℓ λ ( x ) v = x δ ( v ) + λ xv , (61)
r b ( x ) v = v δ ( x ) + b x v . (62) r_b(x)v=v\delta(x)+bxv. \tag*{(62)} r b ( x ) v = v δ ( x ) + b xv . (62)
For b = b α , β b=b_{\alpha,\beta} b = b α , β , write M ( α , β , λ ) = M ( b α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda)=\mathcal{M}(b_{\alpha,\beta},\lambda) M ( α , β , λ ) = M ( b α , β , λ ) . In the basis v i = x i + 1 v_i=x^{i+1} v i = x i + 1 , − 1 ≤ i ≤ p n − 2 -1\leq i\leq p^n-2 − 1 ≤ i ≤ p n − 2 , these actions give (3) –(4) .
Theorem 5.4 (Theorem B). Fix N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) as in Theorem A.
(i) For each λ ∈ k \lambda\in\mathbb{k} λ ∈ k , the bimodule M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) is nontrivial and irreducible of dimension p n p^n p n . Every nontrivial irreducible N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) -bimodule is isomorphic to one of these.
(ii) The N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) -bimodules M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) and M ( α , β , λ ′ ) \mathcal{M}(\alpha,\beta,\lambda') M ( α , β , λ ′ ) are isomorphic if and only if λ = λ ′ \lambda=\lambda' λ = λ ′ .
Proof. We prove the assertions for N ( b ) \mathcal{N}(b) N ( b ) with arbitrary b ∈ A b\in\mathcal{A} b ∈ A , taking b = 1 b=1 b = 1 if n = 0 n=0 n = 0 . Fix λ ∈ k \lambda\in\mathbb{k} λ ∈ k and write E = δ + λ Id E=\delta+\lambda\operatorname{Id} E = δ + λ Id . On M ( b , λ ) = A \mathcal{M}(b,\lambda)=\mathcal{A} M ( b , λ ) = A , the operators are ℓ λ ( x ) = m x E \ell_{\lambda}(x)=m_xE ℓ λ ( x ) = m x E and r b ( x ) = m D ( x ) r_b(x)=m_{D(x)} r b ( x ) = m D ( x ) . Since x ∘ y = x D ( y ) x\circ y=xD(y) x ∘ y = x D ( y ) , one has
ℓ λ ( x ∘ y ) = m x D ( y ) E = r b ( y ) ℓ λ ( x ) , [ r b ( x ) , r b ( y ) ] = 0. \ell_{\lambda}(x\circ y)=m_{xD(y)}E=r_b(y)\ell_{\lambda}(x),\qquad[r_b(x),r_b(y)]=0. ℓ λ ( x ∘ y ) = m x D ( y ) E = r b ( y ) ℓ λ ( x ) , [ r b ( x ) , r b ( y )] = 0.
Identities (56) and (57) follow from
[ m x E , m y E ] = m x δ ( y ) − y δ ( x ) E = ℓ λ ( x ∘ y − y ∘ x ) , [m_xE,m_yE]=m_{x\delta(y)-y\delta(x)}E=\ell_{\lambda}(x\circ y-y\circ x), [ m x E , m y E ] = m x δ ( y ) − y δ ( x ) E = ℓ λ ( x ∘ y − y ∘ x ) ,
[ m x E , m D ( y ) ] = m x δ ( D ( y ) ) = m D ( x D ( y ) ) − D ( y ) D ( x ) = r b ( x ∘ y ) − r b ( y ) r b ( x ) . \begin{aligned}
[m_xE,m_{D(y)}]&=m_{x\delta(D(y))}\\
&=m_{D(xD(y))-D(y)D(x)}=r_b(x\circ y)-r_b(y)r_b(x).
\end{aligned} [ m x E , m D ( y ) ] = m x δ ( D ( y )) = m D ( x D ( y )) − D ( y ) D ( x ) = r b ( x ∘ y ) − r b ( y ) r b ( x ) .
Thus M ( b , λ ) \mathcal{M}(b,\lambda) M ( b , λ ) is a bimodule. Its right actions generate all multiplication operators on A \mathcal{A} A , so it is nontrivial and every subbimodule is an ideal of A \mathcal{A} A . Such an ideal is stable under ℓ λ ( 1 ) = δ + λ Id \ell_{\lambda}(1)=\delta+\lambda\operatorname{Id} ℓ λ ( 1 ) = δ + λ Id and hence under δ \delta δ . The δ \delta δ -simplicity of A \mathcal{A} A proves irreducibility.
Conversely, let V V V be a nontrivial irreducible N ( b ) \mathcal{N}(b) N ( b ) -bimodule, with π \pi π and T = ℓ ( 1 ) T=\ell(1) T = ℓ ( 1 ) as in Lemma 5.2. Write a v = π ( a ) v av=\pi(a)v a v = π ( a ) v . Since V V V is finite-dimensional by Lemma 5.3 and k \mathbb{k} k is algebraically closed, we may choose 0 ≠ w ∈ V 0\ne w\in V 0 = w ∈ V with T w = λ w Tw=\lambda w T w = λ w for some λ ∈ k \lambda\in\mathbb{k} λ ∈ k , following the eigenvector method in [23 Proposition 3.1] and [25 proof of Theorem 4.3, Case 1] . For a ∈ A a\in\mathcal{A} a ∈ A ,
T ( a w ) = ( δ ( a ) + λ a ) w . T(aw)=(\delta(a)+\lambda a)w. T ( a w ) = ( δ ( a ) + λa ) w .
Thus A w \mathcal{A}w A w is a nonzero subspace stable under A \mathcal{A} A and T T T , so V = A w V=\mathcal{A}w V = A w by irreducibility. The map A → V \mathcal{A}\to V A → V , a ↦ a w a\mapsto aw a ↦ a w , is also injective: its kernel is an ideal of A \mathcal{A} A , and if a w = 0 aw=0 a w = 0 , the displayed identity gives δ ( a ) w = 0 \delta(a)w=0 δ ( a ) w = 0 . The kernel is therefore δ \delta δ -stable and proper, so it is zero by δ \delta δ -simplicity. The resulting linear isomorphism F : A → V F:\mathcal{A}\to V F : A → V , F ( a ) = a w F(a)=aw F ( a ) = a w , intertwines both actions: for x , a ∈ A x,a\in\mathcal{A} x , a ∈ A ,
ℓ ( x ) F ( a ) = x T ( a w ) = F ( x δ ( a ) + λ x a ) , \ell(x)F(a)=xT(aw)=F(x\delta(a)+\lambda xa), ℓ ( x ) F ( a ) = x T ( a w ) = F ( x δ ( a ) + λ x a ) ,
r ( x ) F ( a ) = ( δ ( x ) + b x ) a w = F ( ( δ ( x ) + b x ) a ) . r(x)F(a)=(\delta(x)+bx)aw=F((\delta(x)+bx)a). r ( x ) F ( a ) = ( δ ( x ) + b x ) a w = F (( δ ( x ) + b x ) a ) .
Hence V ≅ M ( b , λ ) V\cong\mathcal{M}(b,\lambda) V ≅ M ( b , λ ) and dim V = p n \dim V=p^n dim V = p n .
On M ( b , λ ) \mathcal{M}(b,\lambda) M ( b , λ ) one has
ℓ ( 1 ) p n = ( δ + λ Id ) p n = λ p n Id . (63) \ell(1)^{p^n}=(\delta+\lambda\operatorname{Id})^{p^n}=\lambda^{p^n}\operatorname{Id}. \tag*{(63)} ℓ ( 1 ) p n = ( δ + λ Id ) p n = λ p n Id . (63)
Thus λ \lambda λ is the unique eigenvalue of ℓ ( 1 ) \ell(1) ℓ ( 1 ) . If M ( b , λ ) ≅ M ( b , λ ′ ) \mathcal{M}(b,\lambda) \cong\mathcal{M}(b,\lambda') M ( b , λ ) ≅ M ( b , λ ′ ) , then λ p n = ( λ ′ ) p n \lambda^{p^n}=(\lambda')^{p^n} λ p n = ( λ ′ ) p n and hence λ = λ ′ \lambda=\lambda' λ = λ ′ .
For bimodules V V V over N \mathcal{N} N and V ′ V' V ′ over N ′ \mathcal{N}' N ′ , an isomorphism over an algebra isomorphism ϕ : N → N ′ \phi:\mathcal{N}\to\mathcal{N}' ϕ : N → N ′ is a linear bijection F : V → V ′ F:V\to V' F : V → V ′ satisfying F ( x ∘ v ) = ϕ ( x ) ∘ F ( v ) F(x\circ v)=\phi(x)\circ F(v) F ( x ∘ v ) = ϕ ( x ) ∘ F ( v ) and F ( v ∘ x ) = F ( v ) ∘ ϕ ( x ) F(v\circ x)=F(v)\circ\phi(x) F ( v ∘ x ) = F ( v ) ∘ ϕ ( x ) . For ϕ = Id N \phi=\operatorname{Id}_{\mathcal{N}} ϕ = Id N , this is the usual notion of an isomorphism of N \mathcal{N} N -bimodules.
Corollary 5.5. Let n > 0 n>0 n > 0 .
(i) Let ( σ , h ) ∈ G n (\sigma,h)\in G_n ( σ , h ) ∈ G n , let a 1 a_1 a 1 be the coefficient of x x x in σ ( x ) \sigma(x) σ ( x ) , and put b ′ = ( σ , h ) ∗ b b'=(\sigma,h)*b b ′ = ( σ , h ) ∗ b . Then M ( b , λ ) \mathcal{M}(b,\lambda) M ( b , λ ) and M ( b ′ , a 1 λ ) \mathcal{M}(b',a_1\lambda) M ( b ′ , a 1 λ ) are isomorphic over ϕ = h − 1 σ \phi=h^{-1}\sigma ϕ = h − 1 σ . In particular, G n 1 G_n^1 G n 1 preserves the module parameter.
(ii) Let ϕ c = c − 1 ρ c \phi_c=c^{-1}\rho_c ϕ c = c − 1 ρ c be an algebra isomorphism between N ( α , β ) \mathcal{N}(\alpha,\beta) N ( α , β ) and N ( α ′ , β ′ ) \mathcal{N}(\alpha',\beta') N ( α ′ , β ′ ) as in Theorem 4.4 . The bimodules M ( α , β , λ ) \mathcal{M}(\alpha,\beta,\lambda) M ( α , β , λ ) and M ( α ′ , β ′ , λ ′ ) \mathcal{M}(\alpha',\beta',\lambda') M ( α ′ , β ′ , λ ′ ) are isomorphic over ϕ c \phi_c ϕ c if and only if λ ′ = c λ \lambda'=c\lambda λ ′ = c λ . Every such isomorphism is a nonzero scalar multiple of ρ c \rho_c ρ c , where
ρ c ( v i ) = c i + 1 v i ( − 1 ≤ i ≤ p n − 2 ) . \rho_c(v_i)=c^{i+1}v_i \qquad(-1\leq i\leq p^n-2). ρ c ( v i ) = c i + 1 v i ( − 1 ≤ i ≤ p n − 2 ) .
Proof. (i) On M ( b , λ ) \mathcal{M}(b,\lambda) M ( b , λ ) , define the N ( b ′ ) \mathcal{N}(b') N ( b ′ ) -actions by
ℓ ′ ( a ) = ℓ ( ϕ − 1 ( a ) ) , r ′ ( a ) = r ( ϕ − 1 ( a ) ) . \ell'(a)=\ell(\phi^{-1}(a)),\qquad r'(a)=r(\phi^{-1}(a)). ℓ ′ ( a ) = ℓ ( ϕ − 1 ( a )) , r ′ ( a ) = r ( ϕ − 1 ( a )) .
Put q = p n q=p^n q = p n and g = σ − 1 ( h ) g=\sigma^{-1}(h) g = σ − 1 ( h ) . Since ϕ − 1 ( 1 ) = g \phi^{-1}(1)=g ϕ − 1 ( 1 ) = g , the left action of 1 1 1 becomes T ′ = m g ( δ + λ Id ) T'=m_g(\delta+\lambda\operatorname{Id}) T ′ = m g ( δ + λ Id ) . By the classification for arbitrary b b b established in the proof of Theorem 5.4 , T ′ T' T ′ has a unique eigenvalue μ \mu μ , and the determinant therefore determines the new parameter:
μ q = det T ′ = g ( 0 ) q λ q = a 1 q λ q . \mu^q=\det T'=g(0)^q\lambda^q=a_1^q\lambda^q. μ q = det T ′ = g ( 0 ) q λ q = a 1 q λ q .
Here g ( 0 ) = h ( 0 ) = a 1 g(0)=h(0)=a_1 g ( 0 ) = h ( 0 ) = a 1 , and multiplication by g g g is triangular with diagonal entries g ( 0 ) g(0) g ( 0 ) . Hence μ = a 1 λ \mu=a_1\lambda μ = a 1 λ .
(ii) Taking σ = ρ c \sigma=\rho_c σ = ρ c and h = c h=c h = c in (i) gives the necessary condition λ ′ = c λ \lambda'=c\lambda λ ′ = c λ . Writing b = b α , β b=b_{\alpha,\beta} b = b α , β and b ′ = b α ′ , β ′ b'=b_{\alpha',\beta'} b ′ = b α ′ , β ′ , we have b ′ = c ρ c ( b ) b'=c\rho_c(b) b ′ = c ρ c ( b ) . Together with δ ρ c = c ρ c δ \delta\rho_c=c\rho_c\delta δ ρ c = c ρ c δ , this gives, for x , v ∈ A x,v\in\mathcal{A} x , v ∈ A ,
ℓ c λ ( ϕ c ( x ) ) ρ c ( v ) = c − 1 ρ c ( x ) ( δ ρ c ( v ) + c λ ρ c ( v ) ) = ρ c ( x δ ( v ) + λ x v ) , \ell_{c\lambda}(\phi_c(x))\rho_c(v)=c^{-1}\rho_c(x)(\delta\rho_c(v)+c\lambda\rho_c(v))
=\rho_c(x\delta(v)+\lambda xv), ℓ c λ ( ϕ c ( x )) ρ c ( v ) = c − 1 ρ c ( x ) ( δ ρ c ( v ) + c λ ρ c ( v )) = ρ c ( x δ ( v ) + λ xv ) ,
r b ′ ( ϕ c ( x ) ) ρ c ( v ) = ( ρ c ( δ ( x ) ) + ρ c ( b x ) ) ρ c ( v ) = ρ c ( ( δ ( x ) + b x ) v ) . r_{b'}(\phi_c(x))\rho_c(v)=(\rho_c(\delta(x))+\rho_c(bx))\rho_c(v)
=\rho_c((\delta(x)+bx)v). r b ′ ( ϕ c ( x )) ρ c ( v ) = ( ρ c ( δ ( x )) + ρ c ( b x )) ρ c ( v ) = ρ c (( δ ( x ) + b x ) v ) .
Thus ρ c \rho_c ρ c intertwines both actions over ϕ c \phi_c ϕ c . Composing any such isomorphism with ρ c − 1 \rho_c^{-1} ρ c − 1 on the left gives a bimodule automorphism F F F of M ( b , λ ) \mathcal{M}(b,\lambda) M ( b , λ ) . Since the right actions generate all multiplication operators, F F F commutes with these operators. Put f = F ( 1 ) f=F(1) f = F ( 1 ) . Then
F ( a ) = F ( m a 1 ) = a F ( 1 ) = a f , 0 = [ δ + λ Id , m f ] = m δ ( f ) . F(a)=F(m_a1)=aF(1)=af,\qquad0=[\delta+\lambda\operatorname{Id},m_f]=m_{\delta(f)}. F ( a ) = F ( m a 1 ) = a F ( 1 ) = a f , 0 = [ δ + λ Id , m f ] = m δ ( f ) .
Hence f ∈ ker δ = k 1 f\in\ker\delta=k1 f ∈ ker δ = k 1 , so F F F is a nonzero scalar. Thus all isomorphisms over ϕ c \phi_c ϕ c are nonzero scalar multiples of ρ c \rho_c ρ c . □
Since m h m_h m h commutes with every multiplication operator, it identifies M ( b , γ + κ ) \mathcal{M}(b,\gamma+\kappa) M ( b , γ + κ ) with this bimodule. Thus λ = γ + κ \lambda=\gamma+\kappa λ = γ + κ . For b = b α , β b=b_{\alpha,\beta} b = b α , β , the element s α s_\alpha s α in Proposition 4.7 satisfies L s α = m s α D L_{s_\alpha}=m_{s_\alpha}D L s α = m s α D and s α ( 0 ) = 1 s_\alpha(0)=1 s α ( 0 ) = 1 . Since p n p^n p n is odd, we obtain
κ p n = det D = det L s α = β , λ = γ + β 1 / p n . \kappa^{p^n}=\det D=\det L_{s_\alpha}=\beta,\qquad\lambda=\gamma+\beta^{1/p^n}. κ p n = det D = det L s α = β , λ = γ + β 1/ p n .
In [17 ] , the associative action is recovered through a Novikov–Poisson structure on the split null extension. We use polynomial relations among right multiplication operators (Lemma 5.2 ).
Remark 5.7. Fix n > 0 n>0 n > 0 and a , b ∈ k a,b\in\mathbb{k} a , b ∈ k . Put z = x ( p n − 1 ) z=x^{(p^n-1)} z = x ( p n − 1 ) , u = 1 + b z u=1+bz u = 1 + b z , and ∂ = u δ \partial=u\delta ∂ = u δ . For μ ∈ k \mu\in\mathbb{k} μ ∈ k , consider the actions on O ( 1 ; n ) \mathcal{O}(1;n) O ( 1 ; n )
ℓ μ ( x ) = m x ( ∂ + μ Id ) , r ( x ) = m ∂ ( x ) + a z x . \ell_\mu(x)=m_x(\partial+\mu\operatorname{Id}),\qquad r(x)=m_{\partial(x)}+azx. ℓ μ ( x ) = m x ( ∂ + μ Id ) , r ( x ) = m ∂ ( x ) + a z x .
As in the proof of Theorem 5.4 , these define a nontrivial irreducible X n ( a , b ) \mathcal{X}_n(a,b) X n ( a , b ) -bimodule. Under the normalization ((49) ), the left action of 1 1 1 becomes
T = ℓ μ ( u − 1 ) = δ + m μ − b μ z . T=\ell_\mu(u^{-1})=\delta+m_{\mu-b\mu z}. T = ℓ μ ( u − 1 ) = δ + m μ − b μ z .
In the divided-power basis, T T T has diagonal entries μ \mu μ , entries 1 1 1 immediately above the diagonal, and the additional entry − b μ -b\mu − b μ in the bottom-left corner. Since p n p^n p n is odd, Theorem 5.4 gives
λ p n = det T = μ p n − b μ . (64) \lambda^{p^n}=\det T=\mu^{p^n}-b\mu. \tag*{(64)} λ p n = det T = μ p n − b μ . (64)
Every λ ∈ k \lambda\in\mathbb{k} λ ∈ k occurs, so Theorem 5.4 shows that these exhaust the nontrivial irreducible X n ( a , b ) \mathcal{X}_n(a,b) X n ( a , b ) -bimodules. For fixed a , b a,b a , b , two values μ , μ ′ \mu,\mu' μ , μ ′ give isomorphic bimodules if and only if
( μ − μ ′ ) p n = b ( μ − μ ′ ) . (\mu-\mu')^{p^n}=b(\mu-\mu'). ( μ − μ ′ ) p n = b ( μ − μ ′ ) .
If b ≠ 0 b\ne0 b = 0 , each isomorphism class is represented by exactly p n p^n p n values of μ \mu μ , since the polynomial X p n − b X X^{p^n}-bX X p n − b X has derivative − b -b − b . If b = 0 b=0 b = 0 , then λ = μ \lambda=\mu λ = μ . Our parameter μ \mu μ is denoted by λ \lambda λ in [23 Main Theorem 2, p. 908] . For b ≠ 0 b\ne0 b = 0 , the value μ = 0 \mu=0 μ = 0 is equivalent to any η \eta η satisfying η p n − 1 = b \eta^{p^n-1}=b η p n − 1 = b , so the restriction μ ≠ 0 \mu\ne0 μ = 0 in [23 Main Theorem 2] does not remove an isomorphism class.
Acknowledgments
The authors thank A. P. Pozhidaev for helpful discussions and for sharing the manuscript [17 ] .
Note added. Upon completion of this paper, the authors learned from A. P. Pozhidaev that he and his collaborators had also obtained a classification and an isomorphism criterion for finite-dimensional simple Novikov algebras in an unpublished manuscript.
References
Department of Mathematics, Huizhou Normal University, Huizhou 313000, China Email address: liudong@huznu.edu.cn
Department of Mathematics, Huizhou Normal University, Huizhou 313000, China Email address: pei@huznu.edu.cn
College of Mathematical Sciences, Jiangsu University, Zhenjiang 212013, China Email address: xiali meng@ujs.edu.cn