Do we care about multiple zeta functions? Coming from a number-theoretic background, I certainly care about $L$-functions and in particular automorphic ones. For automorphic forms on $SL_2(\mathbb{Z}) \backslash SL_2(\mathbb{R})$, $L$-function can be interpreted as the Mellin transform of Fourier expansion at the cusp.
If we look at automorphic forms on $SL_(n,\mathbb{Z}) \backslash SL(n,\mathbb{R})$, we still have a Fourier expansion in $n-1$ variables since it is periodic with respect to the super-diagonal unipotent group. This gives us Fourier coefficients $A(m_1,\cdots,m_{n-1})$. In forming the $L$-functions for $SL(n,\mathbb{R})$, we just look at all the Fourier coefficients $A(m,1,\cdots,1)$ and define
$$L(s) = \sum_{m=1}^{\infty} A(m,1,\cdots,1)m^{-s}$$
But the most natural thing to do (comparing to the $SL_2(\mathbb{R})$ case) would be to form the multiple Dirichlet series instead
$$\sum_{m_1,\cdots,m_{n-1}} \frac{A(m_1,\cdots,m_{n-1})}{m_1^{s_1} \cdots m_{n-1}^{s_{n-1}}}$$
It certainly seems that people care "less" about this multiple Dirichlet series: in the whole Langlands business we always take about L-parameters and stuff, which seems to imply that we track only the information of $L$-function but not the whole multiple Dirichlet series.
So here are my questions,


*

*Is there any conceptual reason why we care more about $L$-functions rather than multiple Dirichlet series?

*Of course, Fourier expansion is available when there is a cusp. For cocompact arithmetic quotients of $GL_n(\mathbb{R})$ (or other reductive groups in general), can one similarly define a multiple Dirichlet series that should incorporate the data of $L$-function?


Thank you.
 A: To address the conceptual question, the $L$-function essentially characterizes the automorphic representation and can be studied locally (associating local $L$-functions to local components of the global representation), so it seems to me there is little need (a priori) for using a more complicated Dirichlet series to study these representations.  
On the other hand, multiple Dirichlet series typically do not have Euler products, so do not admit a local-global study, at least in a naive way (though see Bump's survey article).  Of course, as Matt indicates, they are useful for studying $L$-functions.  
Moreover, from an arithmetic point of view, $L$-functions are naturally related to varieties such as elliptic curves.  As far as I know (though I am not an expert on multiple Dirichlet series), there is no direct connection between MDS and counting points on varieties.
A: It may help clarify things to work out a specific example, although the OP may know this.  In case $n=3$, the double Dirichlet series evaluates as
$$ \sum_{m,n=1}^{\infty} \frac{A_F(m,n)}{m^{w} n^s} = \frac{L(\overline{F},w) L(F, s)}{\zeta(s+w)}.$$
Here $\overline{F}$ is the contragredient of $F$.
This is known as Bump's double Dirichlet series, and this is worked out in Section 6.6 of Goldfeld's book, Automorphic forms and $L$-functions for the group $GL(n,\mathbb{R})$.  There are references there for generalizations.
It seems clear (to me) that the $GL_3$ automorphic $L$-function $L(F,s)$ (and its contragredient, and $\zeta$) are the fundamental objects, but also that there are many interesting Dirichlet series that one can construct from an automorphic form.
