Let F be an algebraically closed field of characteristic zero. We consider polynomial maps with constants ω(x,y)=A1xk+A2yk on Mn(F), where A1 is invertible and k≥2. We show that the necessary condition for surjectivity obtained by Saini and Singh (Polynomial maps with constants on matrix algebra, Linear Algebra Appl. 752 (2027), 256-277) is not sufficient in any dimension n≥6. We also completely characterize surjectivity for n=5 and 6.
Introduction
Let F be a field and let A be an associative F-algebra. A polynomial f∈F⟨x,y,…,xm⟩ in non-commuting variables defines a map on A by evaluation. If f is a polynomial identity for A, then its image is {0}. Otherwise, one may ask which elements of A belong to its image. If the map is not surjective, can every element of A be expressed as a sum of a prescribed number of values of f?
A central direction concerns Waring-type problems: whether every element of an algebra can be expressed as a sum of a prescribed number of polynomial values. Representations of matrices as sums of powers form an important class of such problems; see [2, 3, 4, 9]. Related questions concern the structure of polynomial images themselves, including the L’vov–Kaplansky conjecture on images of multilinear polynomials. For an account of polynomial images and Waring-type problems, we refer to [1].
A further variation is obtained by allowing the coefficients of a polynomial to belong to the algebra itself. These polynomials are called generalized polynomials, or polynomials with constants. In this setting, the image depends on the chosen coefficients as well as on the polynomial expression; see [8, 10, 11, 12]. Moreover, the image need not be invariant under conjugation, since conjugating a value also conjugates the coefficients. The study of such maps arises naturally as analogues of word maps with constants in group theory; see [5, 6, 7].
Key words and phrases. Polynomial maps, Polynomial maps with constant, Matrix algebras.
The author acknowledges financial support from the Indian Institute of Science Education and Research (IISER) Mohali through a Postdoctoral Fellowship.
Panja, Saini and Singh [10] studied polynomial maps with constants on M2(F) over an algebraically closed field. In particular, they characterized the surjectivity of maps of the form A1xk1+A2yk2 and proved that the image of A1xy−A2yx is a vector space. Subsequently, Saini and Singh [12] considered the maps
ω:Mn(F)×Mn(F)⟶Mn(F),ω(x,y)=A1xk+A2yk.
where F is algebraically closed of characteristic zero, k≥2, and A1 is invertible. Writing r0=nullity(A2), they proved that n>k(r0−1) is necessary for surjectivity and that this condition is also sufficient for n=3,4. This raises the question of whether the same inequality characterizes surjectivity in higher orders.
We prove that the necessary inequality is sufficient for n=5, but not for n=6. More generally, for every n≥6 and k≥n−1, we characterize surjectivity in terms of the nullity of the second coefficient. Our main results are the following. Throughout, F is algebraically closed of characteristic zero.
Theorem 1.1. Let F be an algebraically closed field of characteristic zero. Let n≥6 and k≥n−1, and A1,A2∈Mn(F) with A1 invertible. Then the map
ω:Mn(F)×Mn(F)⟶Mn(F),ω(x,y)=A1xk+A2yk,
is surjective if and only if nullity(A2)≤1.
Theorem 1.2. Let F be an algebraically closed field of characteristic zero. Let n∈{5,6}, and A1,A2∈Mn(F) with A1 invertible. For an integer k≥2, consider the map
ω:Mn(F)×Mn(F)⟶Mn(F),ω(x,y)=A1xk+A2yk.
Set r0=nullity(A2). Then:
For n=5, the map ω is surjective if and only if 5>k(r0−1).
For n=6, the map ω is surjective if and only if 6>k(r0−1) and (r0,k)=(2,5).
The paper is organized as follows. § 2 contains the reductions and auxiliary lemmas used throughout. In § 3, we establish the general obstruction and deduce the classification for large exponents. § 4 and § 5 give the classifications in orders five and six. In § 6, we conclude with a question concerning higher orders.
Preliminaries
Throughout, we assume F is an algebraically closed field of characteristic 0 and k≥2. We recall the following reductions [12]. Let A1,A2∈Mn(F) be non-zero elements such that one of the coefficients is invertible. Consider the map
ω:Mn(F)2⟶Mn(F),ω(x,y)=A1xk+A2yk.
Without loss of generality, we assume A1 is invertible. Define B=A1−1A2∈Mn(F) and r0=nullity(B)=nullity(A2).
Moreover, the map ω is surjective if and only if the map ωB(x,y)=xk+Byk is surjective. Indeed, ω(x,y)=C if and only if ωB(x,y)=A1−1C.
Reduction to Jordan form. Surjectivity of the map ω remains invariant under the action of the automorphism group of PGLn(F). Thus, we may replace B by its Jordan canonical form
J=i=1⨁r′Jλi,ni⊕(s=1⨁r0J0,ms),
where λi=0 and ∑s=1r0ms=n0. For 1≤s≤r0, set ℓs:=n−n0+m1+⋯+ms, and define σ:{1,…,n}→{1,…,n} by
associated to J and P be the permutation matrix corresponding to σ. Consider the partition of C and J obtained under the conjugation of P:
P−1CP=(C11PC21PC12PC22P)andP−1JP=(J0),
where C11P∈Mn−r0(F), C22P∈Mr0(F) and J∈M(n−r0)×n(F). Note that J is a full row rank. We set L1=(C11PC12P)∈Mn−r0×n(F) and L2=(C21PC22P)∈Mr0×n(F).
Denote W1⊆Fn as the subspace spanned by the first n−r0 rows of CP and W2⊆Fn as the subspace spanned by the last r0 rows of CP. In the following sections, we consider C and B up to conjugation by P.
Notation
Throughout, F denotes an algebraically closed field of characteristic 0, and the split central simple algebra over F is Mn(F), the algebra of all n×n matrices. We denote the automorphism group of Mn(F) by PGLn(F), the projective general linear group, which acts on Mn(F) by conjugation.
For a matrix C∈Mn(F), we denote its rows by c1,…,cn∈Fn and C=(c1,…,cn)⊤. For 1≤i≤n, let ei⊤ denote the i-th standard basis vector of Fn. For C1∈Mn1(F) and C2∈Mn2(F), we define their direct sum by
C1⊕C2=(C100C2)∈Mn1+n2(F).
For B∈Mn(F), we write B∼⨁i=1rJλi,ni, ∑i=1rni=n, to denote a Jordan canonical form of B, where Jλi,ni is the Jordan block of size ni corresponding to the eigenvalue λi. We denote by r0(B) the nullity of B, which turns out to be the number of Jordan blocks corresponding to the eigenvalue 0, and by r′(B) the number of Jordan blocks corresponding to nonzero eigenvalues. Thus, r(B)=r0(B)+r′(B).
For S∈PGLn(F) and C∈Mn(F), we denote SCS−1 by CS. For Z∈Ms×t(F), span(Z) is the span of rows of Z. We retain the above notation throughout the paper.
Auxiliary Lemmas
We first recall the Jordan form of the power of a nilpotent Jordan block.
Lemma 2.1 (Miller). Letn>k≥2. Then the k-th power of the nilpotent Jordan block J0,n is conjugate to
(k−m⨁J0,⌊n/k⌋)⊕(m⨁J0,⌈n/k⌉),
wherem≡n(modk)and0≤m≤k−1.
Lemma 2.2.Forn≥3, every rank-one matrix inMn(F)is a k-th power for2≤k≤n−1.
Proof. Let Z have rank one. If tr(Z)=0, then Z is similar to diag(tr(Z),0n−1) and hence has a k-th root. Otherwise, Z is similar to J0,2⊕0n−2. By Miller’s Theorem 2.1,
J0,2⊕0n−2∼(J0,k+1⊕0n−(k+1))k.
The zero summand is omitted when k=n−1. Therefore, Z=Xk for some X∈Mn(F). □
Lemma 2.3.For1≤s≤n, letR∈Ms×n(F). Ifrank(R)≤n−sorrank(R)=s, then there existH∈Mn−s×n(F)such that the matrix
(HR)=X2for some X∈Mn(F).
Proof. If rank(R)=s, extend its rows to an invertible matrix of order n and take a square root.
Otherwise, set t=rankR≤n−s. For t=0, take X=0. Therefore, we take t=0. Choose Q∈GLs(F) such that the last s−t rows of Q−1R are zero because rank(R)=t<s. Let S=In−s⊕Q and write
Q−1RS={u1,…,ut,0s−t}T.
Since t≤n−s, define
Z={u1,…,ut,0(n−s−t)×n,e1T,…,etT,0(s−t)×n}T.
The last s rows of Z2 are Q−1RS. Let X=SZS−1. Since S=In−s⊕Q, the last s rows of X2=SZ2S−1 are Q(Q−1RS)S−1=R. Therefore, X2 has the required form. □
As a corollary, we obtain:
Corollary 2.4. For s≥1, let R∈Ms×n(F). If rank(R)≤n−s or rank(R)=s, then there exist H∈Mn−s×n(F) such that the matrix (RH)=X2 for some X∈Mn(F).
Proof. Let P=(0s×n−sIn−sIs0n−s×s). Since rank(RP)=rankR, by Lemma 2.3 there exists Z∈Mn(F) such that the last s rows of Z2 are RP. Then (PZP−1)2=PZ2P−1 and hence the first s rows of (PZP−1)2 are therefore RPP−1=R. □
Lemma 2.5. For 1≤s≤n and 2≤k≤n−1, let D∈Ms×n(F) such that rank(D) is s. If H∈Ms×n(F) satisfying rank(H)∈{0,1,s}, then there exists Y∈Mn(F) such that
DYk=H.
Proof. If H=0, take Y=0. Suppose that rank(H)=1. Write H=bw, where b∈Fs is a nonzero column vector and w∈M1×n(F) is nonzero. Since D has full row rank, there exists q∈Fn such that Dq=b. In particular, q=0, so qw has rank one. By Lemma 2.2, there exists Y∈Mn(F) such that Yk=qw. Consequently,
DYk=D(qw)=(Dq)w=bw=H.
Finally, suppose that rank(H)=s. Extend the rows of D and H to bases of Fn, obtaining invertible matrices
S=(DD0),T=(HH0).
With E=(Is0), we have D=ES and H=ET. The invertible matrix S−1T has a k-th root Y∈Mn(F). Hence
DYk=ES(S−1T)=ET=H.
When s=n, the additional rows are omitted. □
Lemma 2.6. Let 1≤r<n, and set s=n−r. Let R∈Mr×n(F) have rank r. For every L∈Ms×n(F), there exists L′∈Ms×n(F) such that N=(L′R) is invertible and
rank(L−L′)=s.
Proof. Extend the rows of R to a basis of Fn, obtaining an invertible matrix
N0=(R0R).
Set E=(Is0) and, for t∈F, define
N(t)=(R0+tER).
Let D be the matrix formed by the first s columns of L−R0. Consider
f(t)=detN(t),g(t)=det(D−tIs).
The polynomial f is nonzero because f(0)=detN0=0, while g has degree s and leading coefficient (−1)s. Since F is infinite, there exists t0∈F such that
f(t0)g(t0)=0.
Taking L′=R0+t0E, we obtain an invertible matrix N=N(t0). Moreover, the first s columns of L−L′ form the invertible matrix D−t0Is. Thus rank(L−L′)=s.
Non-Surjectivity for large exponents
In [12], the authors proved that, when A1 is invertible, the inequality n>k(r0−1), where r0=nullity(A2), is necessary for the surjectivity of ω(x,y)=A1xk+A2yk and is also sufficient for n=3,4. We show below that this condition is not sufficient in general by constructing counterexamples in every order n≥6.
Proposition 3.1. Let F be an algebraically closed field of characteristic zero, let n≥6, and let A∈M(n−2)×n(F) have rank n−2. Set
B=(A02×n).
Then the map Φ(X,Y)=Xn−1+BYn−1 is not surjective.
Proof. Set p=n−2.
We first observe that if Zn−1 has a nontrivial Jordan block at zero, then Z is similar to J0,n, see Theorem 2.1. Indeed, every zero Jordan block of size at most n−1 vanishes upon taking its (n−1)-st power. Consequently,
rank(Zn−1)=1.
Since rankA=p≥4, choose G∈GLp(F) such that the first two rows v1,v2 of GA are linearly independent modulo FenT. Denote the remaining rows by v3,…,vp, and define
R=G−100v1v20(p−4)×n,C=RenT0.
The block 0(p−4)×n in the definition of R is omitted when p=4. We show that C is not in the image of Φ.
Suppose that
C=M+BN,M=Xn−1,N=Yn−1.
The last two rows give
en−1TM=enT,enTM=0.
Thus M has a nontrivial Jordan block at zero, and hence rankM=1. Since its (n−1)-st row is enT, we may write
M=aenTenT0
for some column vector a∈Fp.
Set α=Ga. Multiplying the top-row equation R=aenT+AN on the left by G gives
v1N=−α1enT,v2N=−α2enT,
v3N=v1−α3enT,v4N=v2−α4enT.
The vectors {v3N,v4N} are linearly independent because v1,v2 are independent modulo FenT. Hence rankN≥2.
We next find a nonzero vector z∈span(N) satisfying zN=0. If (α1,α2)=(0,0), the displayed identities imply that enT,v1,v2 belong to span(N). Therefore
z=α2v1−α1v2
is nonzero, belongs to span(N), and satisfies zN=0.
If α1=α2=0, choose (b1,b2)=(0,0) such that b1α3+b2α4=0. Then
z=b1(v1−α3enT)+b2(v2−α4enT)=b1v1+b2v2
is a nonzero element of span(N), and zN=0.
In either case, there exists a row vector w such that wN=z=0 and wN2=0. Thus N has a nontrivial Jordan block at zero. Since N=Yn−1, the initial observation gives rank N=1, a contradiction. □
As a corollary, we obtain the following:
Corollary 3.2. Let n≥6, and let B∈Mn(F) have nullity two. Then the map
(X,Y)⟼Xn−1+BYn−1
is not surjective.
Proof. By similarity, B may be replaced by
(A02×n)
where A has rank n−2. The result follows from Proposition 3.1. □
Proof of Theorem 1.1 Let r0=nullity(A2). If r0≤1, surjectivity follows from [12 Theorem 1.2(1)]. Suppose that r0≥2. If k≥n, then k(r0−1)≥n, so ω is not surjective by [12 Theorem 1.2(2)].
It remains to consider k=n−1. If r0≥3, then k(r0−1)≥2(n−1)≥n, and the same necessary condition gives non-surjectivity. If r0=2, set B=A1−1A2. Since nullity(B)=2, Corollary 3.2 shows that (X,Y)↦Xn−1+BYn−1 is not surjective. Multiplication by A1 gives the required conclusion.
The case n=5
In this section, we prove that the necessary condition obtained in [12] is also sufficient in order five.
Proposition 4.1. Let B∈M5(F) be a non-zero element and k≥2. Define the map
ω:M5(F)×M5(F)⟶M5(F),ω(x,y)=xk+Byk.
The map ω is surjective if and only if 5>k(r0−1).
Proof. The nullity of B is denoted by r0. For r0≤1, surjectivity follows from [12 Theorem 1.2(1)]. For r0≥2, the inequality 5>k(r0−1) is necessary by [12 Theorem 1.2(2)]. Thus, it remains to prove that ω is surjective for 2≤r0≤3 and 2≤k≤4 such that k(r0−1)<5, which is precisely,
(r0,k)∈{(2,2),(3,2),(2,3),(2,4)}.
Denote s=5−r0. By the reduction in § 2, we may assume that
B=(J0),J∈Ms×5(F),rankJ=s.
Let C∈M5(F) be arbitrary element such that
C=(L1L2),L1∈Ms×5(F) and L2∈Mr0×5(F).
Case 1:k=2. Since r0≤3, every possible rank of L2 satisfies one of the hypotheses of Lemma 2.3. Hence there exists L1′∈Ms×5(F) such that
(L1′L2)=X2
for some X∈M5(F). Set H=L1−L1′. Choose U∈GL5(F) satisfying
JU=(0s×r0Is)=E.
Since s≤3, again by Lemma 2.3, there exists Z∈M5(F) such that EZ2=HU. Taking Y=UZU−1, we obtain
JY2=JUZ2U−1=EZ2U−1=H.
Therefore X2+BY2=C.
Case 2:r0=2, k∈{3,4}. By Lemma 2.5, it suffices to construct a k-th power matrix
N=(L1′L2)
such that the rank of L1−L1′ is at most one or s. Indeed, writing N=Xk, the lemma gives Y∈Mn(F) satisfying JYk=L1−L1′, and therefore Xk+BYk=C.
Set E=(I30). If rank(L2)=2, then Lemma 2.6 gives L1′∈Ms×n(F) such that
N=(L1′L2)
is invertible and rank(L1−L1′)=s. Thus N is a k-th power with the required rank.
If L2=0, take N(t)=tI3⊕02. Clearly, N=Xk for X=ktI3⊕02. The determinant of the first three columns of L1−tE has leading term −t3, so a suitable nonzero t0 makes the matrix C−N have rank three.
It remains to consider rankL2=1. Recall that L2=(C21PC22P). If rankC21P=1, then C∈Image(ω) by [12 Proposition 4.5] as dim(W2)=rank(C21P). We may therefore assume that C21P=0 and $\operatorname{rank}C^P_{22}=1. \end_mid} Let Q∈GL2(F) such that QC22PQ−1 is the Jordan canonical form of C22P. Simultaneous conjugation by S:=I3⊕Q preserves the zero lower rows of B and the rank of J^. Putting C22P in Jordan canonical form, reduces L2 to one of
(μe4T0),(e5T0),μ=0.
For the first form, take N(t)=tI3⊕(μ)⊕0, which is a k-th power. A suitable nonzero t0 makes H=L1−t0E have rank three.
For the second form, if rankL1=2, take N=E45. By Lemma 2.2, N is a k-th power, and CS−N is of the form
(L102×5),
having rank at most one or three.
Finally, suppose rankL1=2. Let c1,c2,c3 be its rows, and set
W1=span{c1,c2,c3}.
If e5T∈W1, choose a basis {e5T,w} of W1 and write ci=aie5T+biw for 1≤i≤3. Take
N=a1a2a310e5T.
Then N and C−N both have rank at most one.
If e5T∈/W1, choose independent rows cj,cℓ, and let i be the remaining index. Take
N=(e4+ei)e5T.
The first three rows of C−N are cj,cℓ and ci−e5T, in their original order. They are linearly independent because e5T∈/W1. Hence the matrix C−N has rank three.
In both subcases, N has rank one and is a k-th power by Lemma 2.2. Undoing the simultaneous conjugations completes the proof. □
Proof of Theorem (1)
As a corollary, we present the proof of Theorem 1.2(1):
4.1. Proof of Theorem 1.2(1). Since A1 is invertible, let B=A1−1A2. The nullity(B) = nullity(A2). Therefore, if C=xk+Byk, then A1C=A1xk+A2yk. Thus, ωB is surjective if and only if ω is surjective. Hence, ω is surjective if and only if 5>k(r0−1).
The case n=6
As proved in § 3, the condition n>k(r−1) is necessary but not sufficient for surjectivity. Indeed, for n=6, it holds when r=2 and k=5, whereas the corresponding map is not surjective, as shown in Proposition 3.1. In this section, we prove that for n=6 this is the only exception and obtain the following classification.
We first show that every square-zero nilpotent matrix of rank two in order six has a k-th root for 2≤k≤4.
Lemma 5.1. Let N∈M6(F) such that rank(N)=2 and N2=0. Then there exist X∈M6(F) such that N=Xk for k=2,3,4.
Proof. The Jordan canonical form of N is J0,2⊕J0,2⊕02. For k=2, take X=(J0,4⊕02), for k=3 take X=(J0,5⊕0) and X=J0,6 for k=4. In each case, by Theorem 2.1, Xk is similar to N. □
Proposition 5.2. Let B∈M6(F) be a non-zero element, and k≥2 be an integer. The map ω(x,y)=xk+Byk is surjective on M6(F) if and only if 6>k(r0−1) and (r0,k)=(2,5), where r0=nullity(B).
Proof. For r0≤1, surjectivity follows from [12 Theorem 1.2(1)]. For r0≥2, the condition 6>k(r0−1) is necessary by [12 Theorem 1.2(2)]. The pairs satisfying this condition are precisely
(r0,k)∈{(2,2),(2,3),(2,4),(2,5),(3,2)}.
For (r0,k)=(2,5), the map ω is not surjective by Proposition 3.1. Thus, we prove the surjectivity for the remaining pairs.
Let s=6−r0. By the reduction in § 2, we may assume that B=(J0), where J∈Ms×6(F) and rankJ=s.
Let C∈M6(F) be arbitrary, and we write C=(L1L2), where L1∈Ms×6(F) and L2∈Mr0×6(F).
Case 1:(r0,k)=(3,2). By Lemma 2.3, there exists X∈M6(F) such that X2=(L1′L2) for some L1′∈M3×6(F). Set H=L1−L1′. Choose U∈GL6(F) satisfying
JU=(03×3I3)=E.
Since rank(HU)≤3, by Lemma 2.3, there exists Z∈M6(F) such that EZ2=HU. Taking Y=UZU−1 gives
JY2=JUZ2U−1=H,
and hence X2+BY2=C.
Case 2:r0=2, 2≤k≤4. By Lemma 2.5, it suffices to construct a k-th power
N=(L1′L2)
such that
rank(L1−L1′)∈{0,1,s}.
Indeed, writing N=Xk, the lemma gives Y∈Mn(F) satisfying J^Yk=L1−L1′, and therefore Xk+BYk=C.
Set E=(I40). If rank(L2)=2, then Lemma 2.6 gives L1′∈Ms×n(F) such that
N=(L1′L2)
is invertible and rank(L1−L1′)=s. Thus N is a k-th power with the required rank.
If L2=0, take N(t)=(tI4⊕02)=(ktI4⊕02)k. A suitable nonzero t0 makes L1−t0E have rank four.
It remains to consider rank(L2)=1. Recall Section 2, L2=(C21PC22P).
If rankC21P=1, then C belongs to the image by [12 Proposition 4.5]. We may therefore assume that C21P=0 and rankC22P=1.
A simultaneous conjugation by I4⊕Q, with Q∈GL2(F), preserves the zero lower rows of B and the rank of J^. Therefore, consider C22P in Jordan canonical form and thus L2 reduces to one of
(μe5T0),(e6T0),μ=0.
For the first form, take N(t)=tI4⊕(μ)⊕0=Xk for X=ktI4⊕(kμ)⊕0. We can choose t0 such that H=L1−t0E has rank four.
Now, we consider the second form, C=L1e6T0. If rankL1∈{0,1,4}, take N=E56, which is a k-th power by Lemma 2.2 and C−N=(L10).
Suppose that rank L1=2. If e6T∈W1, choose a basis {e6T,w} of W1 and write the rows of L1 as ci=aie6T+biw for 1≤i≤4. Then
Thus, C is the sum of two rank-one matrices. The matrix N=Xk for some X by Theorem 2.2 and C−N is again a rank-one matrix.
If e6T∈/W1. Since rank(L1)=2, there exist Q∈GL4(F) such that QL1=u1u200. Let S=Q⊕I2. Simultaneous conjugation of C=Xk+Jyk by S gives
QL1e6T0=Xk+(QJJS−102×6)yk.
where QL1S−1=u1S−1u2S−100. The conjugation by S preserves the last two rows of C, the form of B, and the condition e6T∈/W1.
Consider V=span{e1T,e2T,e3T−e5T}. Here W1 denotes the row space of the transformed top block QL1S−1. Both V and W1+Fe6T have dimension three. However, e6T∈W1+Fe6T whereas e6T∈/V. Hence these subspaces are distinct, and we may choose w∈V∖(W1+Fe6T). Let N=(0,0,e6T,w,e6T,0)T. Since w∈V, we have w=(a,b,c,0,−c,0), we obtain wN=0. Also e6TN=0, so N2=0. The row space of N is spanned by w and e6T, so rankN=2. By Lemma 5.1, N is a k-th power.
Thus, we have
CS−N=u1S−1u2S−1−e6T−w02×6.
Proof of Theorem (2)
The first two rows span W1. Since e6T∈/W1 and w∈/W1+Fe6T, the first four rows are linearly independent. Thus rank(CS−N)=4.
Finally, suppose that rankL1=3. If e6T∈/W1, choose three independent rows of L1, and let i be the remaining index. Taking N=(e5+ei)e6T gives a rank-one k-th power such that C−N has rank four.
If e6T∈W1, a similar construction as above allows us to assume that L1=u1u2u30 where u1,u2,u3 are independent. Let V=span{e1T,e2T,e3T}. Since V and W1 have dimension three and e6T∈W1∖V, choose w∈V∖W1. Define N=(0,0,0,w,e6T,0)T. Now, wN=0 and e6TN=0, and thus N2=0. Moreover, rankN=2, so N is a k-th power by Lemma 5.1. The matrix C−N has rank four, with the last two rows being zero.
In every case, we have constructed N=Xk with the prescribed last two rows and rank(L1−L1′)∈{0,1,4}.
5.1. Proof of Theorem 1.2(2). Since A1 is invertible, let B=A1−1A2. The nullity(B) = nullity(A2). Therefore, if C=Xk+BYk, then A1C=A1Xk+A2Yk. Hence, ω is surjective if and only if 6>k(r0−1) and (r0,k)=(2,5).
Concluding remarks
The results above complete the surjectivity classification in orders five and six and show that the necessary condition n>k(r0−1) is not sufficient in any order n≥6. For coefficients of nullity two, the case k=n−1 is settled by the general obstruction. This leaves the following question for smaller exponents.
Question 6.1. Let n≥7 and 2≤k≤n−2. For which matrices B∈Mn(F) of nullity two is the map
ωB:Mn(F)×Mn(F)⟶Mn(F),ωB(X,Y)=Xk+BYk,
surjective?
References
Indian Institute of Science Education and Research Mohali, Knowledge City, Sector 81, Punjab 140306, India
[1]M. Brešar and C. Martínez. Waring Problems Across Algebra. 2026. arXiv: 2603. 07033.
[2]M. Brešar and P. Šemrl. The Waring problem for matrix algebras. In: Israel J. Math. 253.1 (2023), pp. 381–405. ISSN: 0021-2172,1565-8511. DOI: 10 . 1007/ s11856-022-2366-7.arxiv.org/abs/2103.10708
[3]M. Brešar and P. Šemrl. The Waring problem for matrix algebras, II. In: Bull. Lond. Math. Soc. 55.4 (2023), pp. 1880–1889. DOI: 10.1112/blms.12825.DOI
[4]I. G. Gargate and T. C. de Mello. Images of multilinear polynomials on n × n upper triangular matrices over infinite fields. In: Israel J. Math. 252.1 (2022), pp. 337–354. DOI: 10.1007/s11856-022-2350-2.DOI
[5]F. A. Gnutov. On the image of a word map with constants of a simple algebraic group II. In: Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 492 (2020). MR4223709, pp. 75–93.
[6]N. Gordeev, B. Kunyavskiĭ, and E. Plotkin. Word maps, word maps with constants and representation varieties of one-relator groups. In: J. Algebra 500 (2018), pp. 390–424. DOI: 10.1016/j.jalgebra.2017.03.016.DOI
[7]M. Larsen and A. Shalev. Word maps and Waring type problems. In: J. Amer. Math. Soc. 22 (2009), pp. 437–466. DOI: 10.1090/S0894-0347-08-00615-2.DOI
[8]T. C. D. Mello and P. Saini. Generalized Commutator on Octonion Algebra. 2026. arXiv: 2610.04466 [math.RA].arxiv.org/abs/2610.04466
[9]S. Panja and S. Prasad. The image of polynomials and Waring type problems on upper triangular matrix algebras. In: J. Algebra 631 (2023), pp. 148–193. DOI: 10.1016/j.jalgebra.2023.04.027.DOI
[10]S. Panja, P. Saini, and A. Singh. Images of polynomial maps with constants. In: Mathematika 71 (2025), Paper No. e70031 (17 pages). DOI: 10.1112/mtk.70031.DOI
[11]S. Panja, P. Saini, and A. Singh. Polynomial maps with constants on split octonion algebras. In: Comm. Algebra 54 (2026), pp. 579–594. DOI: 10.1080/00927872. 2025.2531559.
[12]P. Saini and A. Singh. Polynomial maps with constants on matrix algebra. In: Linear Algebra and its Applications 752 (2027), pp. 256–277. ISSN: 0024-3795. DOI: 10.1016/j.laa.2026.08.016.DOI