On Multiple Eisenstein Series in Positive Characteristic: Direct Sum Result

By (?), for every s\geq1 with q-1\mid s, one checks that

and hence

By Proposition 2.7, we have an equality of graded L-algebras

Thus, if F\in\mathscr{M}^{(r)}_{L,w}, then F is an L-linear combination of monomials in E_{q-1}^{(r)},\ldots,E_{q^r-1}^{(r)} of total weight w. Since each E_{q^i-1}^{(r)} belongs to \mathcal{Z}_{L,q^i-1}^{(r)}, it follows from (?) that

The first assertion follows by summing over all weights.

For r=1, the assertions follow from Remark 2.9 and the fact that \phi^A:A\to\mathbb{C}_\infty[\tau] is an \mathbb{F}_q-algebra homomorphism. For r\geq 2, see [BBP24, Propositions 15.12 and 16.4].

Let d=\deg f. We note that from (?) and (?), we have

By Lemma 3.1, together with Remark 2.9 for r=1 and Lemma 2.8 for r\geq2, we have

and hence

Expanding the inverse as a formal power series gives

Consider a monomial occurring on the right-hand side,

Its coefficient belongs to \mathcal{L}_{K, \sum_{j=1}^m(q^{i_j}-q^{rd})}^{(r)}, while its power of t_{\Lambda_{\mathbf{z}'}}(z_1) is

Thus, if such a monomial contributes to the coefficient of t^N in t_{\Lambda_{\mathbf{z}'}}(fz_1), then

and its coefficient belongs to

Therefore every term contributing to \mathcal{A}_{f,N}(\mathbf{z}') lies in \mathcal{L}^{(r)}_{K, 1-N}, and hence

for all N\geq0.

Recall that the coefficient e_n(\mathbf{w}) \in \mathcal{O}(\Omega^r) of \exp_{\Lambda_{\mathbf{w}}}(X) is defined in (?). By Remark 2.9 for r=1 and Proposition 2.7, Lemma 2.8 for r\geq2, we have

According to [Gekeler88, (3.8)], the Goss polynomial associated with \Lambda_{\mathbf{w}} is given by

where the inner sum runs over all tuples \underline{i}=(i_0,\ldots,i_{s-1}) of non-negative integers satisfying

and

Hence, by (?),

Taking j=i-1, we obtain

Using the same notation as in Lemma 3.2 and Lemma 3.3, we see that

Comparing the coefficient of t_{\Lambda_{\mathbf{z}'}}(z_1)^N, we obtain

By Lemma 3.2 and Lemma 3.3, for each 1\le i \le s, we have

Therefore, \mathcal{C}_{s,f,N}(\mathbf{z}') \in \mathcal{L}_{K,s-N}^{(r)}, as desired.

We may assume that \mathfrak{s}=(s_1,\ldots,s_m) is nonempty. Fix f_1,\ldots,f_m\in A_+ with \deg f_1 > \cdots > \deg f_m. Then by Lemma 3.4, we have

and

Now, note that by (?), for all s\geq 1 and f\in A_+, we have

In particular, if \deg f > \frac{1}{r}\log_q (N+1), then \operatorname{ord}_{r+1}(G_s(t_{\Lambda_{\mathbf{z}'}}(fz_1))) > N. Hence, for all N\geq 0,

Therefore, we obtain

We may assume that \mathfrak{s}=(s_1,\ldots,s_m) is nonempty. By Proposition 2.6, we have

For each 0\leq i\leq m, write

By Proposition 3.5, we have

Moreover,

Hence

Comparing the coefficients of t_{\Lambda_{\mathbf{z}'}}(z_1)^N, we obtain

as desired.

Suppose that

where F_i\in\mathcal{Z}_{\mathbb{C}_\infty,i}^{(r)} for every 0\leq i\leq w. For each i, choose an expression

where c_{i,j}\in\mathbb{C}_\infty and \operatorname{wt}(\mathfrak{s}_{i,j})=i, and put

By Proposition A.6, as N\to\infty,

Evaluating the given relation at \mathbf{z}_N and applying Proposition A.6 again, we obtain

Dividing by j_N(\mathbf{z})^w yields

Since F_i^{\langle N \rangle}(\mathbf{z})\to F_i(\mathbf{z}) as N\to\infty, the sequences \{F_i^{\langle N \rangle}(\mathbf{z})\}_{N\geq1} are bounded. Moreover, for N\gg0,

Hence, letting N\to\infty in the preceding identity yields

Repeating the same argument for F_0+\cdots+F_{w-1}=0, we obtain

Therefore,

It suffices to show the case r=2. For general r\geq 2, the result follows by adapting the induction argument of Theorem 1.5 (ii), with \overline{K} replaced by \mathbb{C}_\infty and the base case of the induction replaced by the case r=2.

Assume r=2 and w\ge 1. The map \widetilde{\mathcal{Z}}_{K,w}^{(2)}\otimes_K \mathbb{C}_\infty \to \widetilde{\mathcal{Z}}_{\mathbb{C}_\infty,w}^{(2)} fits into the following commutative diagram:

where the three isomorphisms follow from Proposition A.8 and (?). This implies \widetilde{\mathcal{Z}}_{K,w}^{(2)}\otimes_K \mathbb{C}_\infty \xrightarrow{\sim} \widetilde{\mathcal{Z}}_{\mathbb{C}_\infty,w}^{(2)} is an isomorphism. Therefore, by (?), we obtain

The assertion for r=2 follows from summing over all w\ge 0. This completes the proof.

If G_2(\mathfrak{s}; \mathbf{z}) = 0, then

By Theorem A.1, this implies

On the other hand, by [CCHT26, Proposition 2.10], the constant term of the t_{\Lambda_{\mathbf{z}'}}(z_1)-expansion of E_2(\mathfrak{s};\mathbf z) is E_1(\mathfrak{s})=\zeta_A(\mathfrak{s}), which is nonzero by [Tha09, Theorem 4]. This is a contradiction.

Write \bm{f}=(f_1,\ldots,f_r). We have

If \ell(\bm{f})\leq r-2, the first nonzero coordinate of \bm{f}\gamma_N is the same as that of \bm{f}, so the first assertion is immediate. If \ell(\bm{f})=r-1, then \deg f_r<N, while

Hence \theta^Nf_{r-1}+f_r is monic if and only if f_{r-1} is monic. Finally, if \ell(\bm{f})=r, then

so again \bm{f}\succ_N0 if and only if \bm{f}\succ0. This proves (i). The proof of (ii) follows from a similar argument.

We may assume \mathfrak{s}=(s_1,\ldots,s_m) is nonempty. By Lemma A.4, the two summation sets defining E_r^{\langle N\rangle}(\mathfrak{s};\mathbf{z}) and E_r(\mathfrak{s};\mathbf{z}) agree whenever

Hence their difference is supported on tuples for which |\bm{f}_i|\geq q^N for some 1\leq i\leq m. On the other hand, notice that for each fixed \mathbf{z}\in\Omega^r, the K_\infty-linear map

is injective. Since K_\infty^r is finite-dimensional, there exists c_0(\mathbf{z})>0 such that

Thus, a standard estimate shows that there exists a constant c(\mathbf{z})>0, independent of N, such that

This proves (i).

For (ii), one checks that

which implies \mathbf{z}_{N}\in \Omega^{r}. It follows that

where the fourth equality follows by the change of variables \bm{f}_i\mapsto\bm{f}_i\gamma_N^{-1} and the definition of \succ_N. This proves (ii).

Notice that for any \mathbf{z}\in \Omega^2, \Lambda_{\mathbf{z}'}=A. Let C=\phi^{\widetilde{\pi}A} denote the Carlitz A-module and \exp_C(X)=\exp_{\widetilde{\pi}A}(X)\in K[\![X]\!] denote the Carlitz exponential (see [goss1996basic, §3]). By Definition 2.2, Proposition 2.3, we have

On the other hand, for every f\in A_+, we have

It follows that for every nonempty index \mathfrak{s}=(s_1,\ldots,s_m),

Here, for each M\geq 0, (?) shows that only finitely many terms in the series above contribute to its t_{\widetilde{\pi}A}(\widetilde{\pi}z_1)^M-coefficient. Hence this coefficient belongs to K. Therefore, we get an injective K-linear map

Since \mathbb{C}_\infty is flat over K,

is still an injection. Notice that the image \widetilde{\mathcal{G}}_{K,\le w}^{(2)}\otimes_K \mathbb{C}_{\infty}\to \mathbb{C}_{\infty}[\![t_{\widetilde{\pi}A}(\widetilde{\pi}z_1)]\!] is \widetilde{\mathcal{G}}_{\mathbb{C}_{\infty},\le w}, so the assertion follows.

Assume w\geq1 as the case w=0 is clear. We first show the well-definedness for L = \mathbb{C}_{\infty}. Suppose

for some c_i\in \mathbb{C}_\infty. Then by (?) we have

Hence, \varphi_{\mathbb{C}_{\infty}, w} is well-defined.

Now, we show the well-definedness for L=K. Suppose that

for some c_i\in K. Similarly, by (?) and (?), we see that

Hence, \varphi_{K, w} is well-defined.

For L=K or L=\mathbb{C}_{\infty}, we note that the surjectivity of \varphi_{L,w} is clear from the definition. We then show the injectivity of \varphi_{L, w} as follows. Suppose that \sum_{i=1}^{m}c_i\widetilde{E}_2(\mathfrak{s}_i; \mathbf{z})\in \widetilde{\mathcal{Z}}^{(2)}_{L, w} such that

By (?), we have

which implies

Therefore, by Theorem A.1 (i), we obtain

This proves the injectivity.