The spectral function of the Tomonaga-Luttinger model revisited: power laws and universality
Extracted source text ยท arXiv 1509.06861
%\documentclass[prl,twocolumn,showpacs]{revtex4}
\documentclass[prb, twocolumn, showpacs, amsmath, amssymb]{revtex4-1}
%\bibliographystyle{apsrev4-1}
\usepackage{amssymb,amsfonts,amsmath}
\usepackage{graphicx}% Include figure files
%\usepackage{bm}% bold math
\usepackage{color}
\usepackage{hyperref}
\usepackage{longtable}
\newcommand{\ix}[1]{\ensuremath{\text{#1}}} % for indices in upright
\DeclareMathOperator{\Tr}{Tr} % trace
\newcommand{\K}{\ix{K}} % Keldysh index
\newcommand{\Ret}{\ix{ret}} % Ret-index
\newcommand{\Adv}{\ix{adv}} % Adv-index
\usepackage{bbm}
\begin{document}
\title{The spectral function of the Tomonaga-Luttinger model revisited: \\ power laws and universality}
\author{L.\ Markhof}
\affiliation{Institut f{\"u}r Theorie der Statistischen Physik, RWTH Aachen University
and JARA---Fundamentals of Future Information
Technology, 52056 Aachen, Germany}
\author{V.\ Meden}
\affiliation{Institut f{\"u}r Theorie der Statistischen Physik, RWTH Aachen University
and JARA---Fundamentals of Future Information
Technology, 52056 Aachen, Germany}
\begin{abstract}
We reinvestigate the momentum-resolved single-particle spectral function of the Tomonaga-Luttinger model.
In particular, we focus on the role of the momentum-dependence of the two-particle interaction $V(q)$.
Usually, $V(q)$ is assumed to be a constant and integrals are regularized in the ultraviolet
`by hand' employing an ad hoc procedure. As the momentum dependence of the interaction is irrelevant in
the renormalization group sense this does not affect the universal low-energy properties of the model,
e.g. exponents of power laws, if {\it all} energy scales are sent to zero.
If, however, the momentum $k$ is fixed away from the Fermi momentum $k_{\rm F}$, with
$|k-k_{\rm F}|$ setting a nonvanishing energy scale, the details of $V(q)$ start to matter.
We provide strong evidence that any curvature of the two-particle interaction at small
transferred momentum $q$ destroys power-law scaling of the momentum resolved spectral function
as a function of energy. Even for $|k-k_{\rm F}|$ much smaller than the momentum space range
of the interaction the spectral line shape depends on the details of $V(q)$. The significance
of our results for universality in the Luttinger liquid sense, for experiments on quasi
one-dimensional metals, and for recent results on the spectral function of
one-dimensional correlated systems taking effects of the curvature of the single-particle
dispersion into account (`nonlinear LL phenomenology') is discussed.
\end{abstract}
\pacs{}
\date{\today}
\maketitle
\section{Introduction}
\subsection{Luttinger liquid universality and the Tomonaga-Luttinger model}
It is well established that the Tomonaga-Luttinger model (TLM)\cite{Tomonaga50,Luttinger63}
with linear single-particle dispersion and a two-particle interaction potential
$V(q)$ which is finite at zero momentum
transfer $q=0$ forms the infrared fixed point under renormalization group (RG) flow of a
large class of gapless one-dimensional (1d) models of correlated fermions.\cite{Solyom79}
This is the essence of the much celebrated Luttinger liquid (LL)
universality.\cite{Haldane81,Voit95,Giamarchi03,Schoenhammer05}
It implies that the low-temperature thermodynamic properties as well as the
low-energy spectral functions of a model belonging to the LL universality class are
equivalent to the ones of the TLM. Understanding the low-energy physics of the latter is
thus of crucial importance. Fortunately, using bosonization\cite{Haldane81,vonDelft98,Schoenhammer05}
it is possible to derive exact and closed analytical expressions for thermodynamic
observables such as the specific heat or the compressibility as well as for space-time
correlation functions of the TLM. From the latter spectral functions can be computed by
Fourier transform.
The bosonization expressions for correlation functions of the TLM depending
on position $x$ and time $t$ generically
contain integrals over momenta. Within constructive bosonization, which is based on operator
identities,\cite{Haldane81,Schoenhammer05} these are naturally regularized in the
ultraviolet by the momentum-space range $q_{\rm c}>0$ of the two-particle potential $V(q)$.
For time-dependent correlation functions
the momentum integrals cannot be performed even if a specific form of $V(q)$ is assumed.\cite{Meden99}
However, one can show that the momentum dependence of the interaction is RG irrelevant.\cite{Solyom79}
This is employed to justify the following procedure: In the final expressions for the space-time correlation
functions $V(q)$ is routinely replaced by a constant. As a consequence the momentum integrals become
divergent in the ultraviolet. These divergences are regularized (`by hand') applying an ad hoc
procedure.\cite{Luther74,Voit95,Giamarchi03} We already now emphasize that this regularization is not unique.
After these steps the momentum integrals can be performed
and integral-free expressions for space-time correlation functions are obtained. In field-theory
inspired phenomenological bosonization procedures \cite{Giamarchi03} the momentum dependence of
the interaction is often neglected from the outset (even in the Hamiltonian). In correlation
functions this leads to the same ultraviolet divergences as described above requiring again
a regularization `by hand'. Similarly, the purely fermionic approach
to the single-particle Green function $G^\gtrless(x,t)$ of the TLM by Dzyaloshinski\v{i}
and Larkin \cite{Dzyaloshinskii73} requires an ad hoc ultraviolet
regularization.
We thus emphasize that the integral-free expressions for a variety of
time-dependent correlation functions which can be found in the literature
cannot be considered as the corresponding {\it exact} correlation functions of the
TLM. This is often acknowledged by stating that the integral-free expressions of the ad
hoc procedure only agree to the exact ones at `asymptotically large space-time
distances'; as discussed in Ref.~\onlinecite{Meden99} (see also below), even this is
incorrect when considering the decay in the directions specified by
$x = \pm v t$,
with $v$ being one of the eigenmode velocities at small momentum.
\subsection{Spectral functions of the Tomonaga-Luttinger model -- the fate of power laws}
\label{sect_fate}
We now focus on the two-point correlation function -- the Green function --
at temperature $T=0$ from which the single-particle spectral function can be computed by
Fourier transform.
The spectral function is of particular importance as it provides forthright access to
correlation effects and can directly be related to photoemission spectra. We consider the momentum
integrated function $\rho^{<}(\omega)$ [$\rho^{>}(\omega)$], which is experimentally
accessible in angular integrated [inverse] photoemission, as well as the momentum resolved
spectral function $\rho^{<}(k,\omega)$ [$\rho^{>}(k,\omega)$]. A measurement of the latter
requires momentum resolution. It was shown that the universal low-energy
power-law suppression of $\rho^\gtrless(\omega) \sim |\omega|^\alpha$ for $\omega \to 0$, with
$\alpha>0$ is unaffected by the above ad hoc procedure.\cite{Meden99} Regardless of the
details of $V(q)$ the exponent $\alpha$ depends on the potential at vanishing momentum
transfer $V(0)$ (only), that is the constant interaction strength after the ad hoc
procedure. Furthermore, the power-law nonanalyticity of $\rho^\gtrless(k_{\rm F},\omega)
\sim |\omega|^{\alpha-1}$ exactly at the Fermi momentum $k=k_{\rm F}$ remains the same; depending
on the size of $V(0)$ a divergence ($\alpha<1$) or suppression ($\alpha>1$) might
occur. These findings are consistent with the RG irrelevance of the momentum dependence
of $V(q)$ as in both cases {\it all} energy scales, that is $\omega$ {\it and,} in the
case of momentum resolved spectra, $v_{\rm F} (k-k_{\rm F})$ are sent to zero.\cite{Solyom79}
Here $v_{\rm F}$ denotes the Fermi velocity.
The question we address here is whether or not any of the standard ad hoc procedures is
legitimate when it comes to $\rho^\gtrless(k,\omega)$ as a function of $\omega$ at fixed $k-k_{\rm F} \neq 0$.
Employing these to compute $\rho^\gtrless(k,\omega)$ of the spinless TLM, characteristic
algebraic threshold nonanalyticities at $\pm v_{c} [k-k_{\rm F}]$ with the charge
velocity $v_{c}$ of the (collective, bosonic) charge eigenmodes were
found. In the model with spin
additional algebraic nonanalyticities appear at $\pm v_{s} [k-k_{\rm F}]$ with the
velocity $v_{s}$ of the spin modes.\cite{Theumann67,Dzyaloshinskii73,Luther74,Meden92,Voit93}
The corresponding exponents
can be expressed in terms of the (momentum independent) interaction potential. For
small interactions and $k-k_F<0$, $\rho^{<}(k,\omega)$ shows {\it power-law singularities}
at $\omega = v_{c} [k-k_{\rm F}]$ and $\omega = v_{s} [k-k_{\rm F}]$ instead of a
single (Lorentzian) peak which would emerge in a Fermi liquid. This is one of the signatures of
spin-charge separation regarded as a hallmark of (spinful) LLs. In Ref.~\onlinecite{Schoenhammer93} it
was shown that these features can be found in the exact spectral function of the TLM if a box-like
potential $V(q) = V(0) \Theta(q_{\rm c}^2 - q^2)$ is assumed, as long as $|k-k_{\rm F}| < q_{\rm c}$
with the momentum transfer cutoff $q_{\rm c}$. We emphasize that using a box potential is
per se not equivalent to the ad hoc procedure, as it can e.g. be seen considering
$\rho^\gtrless(k,\omega)$ for $|k-k_{\rm F}| > q_{\rm c}$.\cite{Schoenhammer93}
For $|k-k_{\rm F}| < q_{\rm c}$ the box potential might, however,
be viewed as a unique realization of the ad hoc procedure (see below).
Clearly, a box potential is rather special and might not even be considered as particular
physical. Thus further work for more generic $V(q)$ is required.
In Ref.~\onlinecite{Meden99} it was shown that the algebraic properties of the
Green functions $G^\gtrless(x,t)$ in the space-time plane are significantly affected by the ad hoc procedures.
It was proven that the exponent of the asymptotic decay of the Green function in the distinguished
directions $x = \pm v_{c/s} t$ is not only set by $V(0)$ but in addition by a measure
of the flatness of the potential at $q \to 0$; a result which cannot be obtained within
any ad hoc procedure. Based on this and the crucial insight that the decay of $G^\gtrless(x,t)$
in the distinguished directions plays a central role in obtaining the power-law
nonanalyticities in $\rho^\gtrless(k,\omega)$, the question was posed if for generic $V(q)$, $\rho^\gtrless(k,\omega)$ is characterized
by the `thought to be universal' power laws of the ad hoc procedure. However,
Ref.~\onlinecite{Meden99} lacks a definite answer.
The above mentioned box potential is
`infinitely flat' at $q \to 0$ and thus `nongeneric'. The asymptotics of $G^\gtrless(x,t)$
for this and the ad hoc procedures agree and consequently also the features of $\rho^\gtrless(k,\omega)$
at $\omega \approx v_{c/s} [k-k_{\rm F}]$.
It is crucial to realize that a dependence of the line shape of $\rho^\gtrless(k,\omega)$ on the details
of the interaction away from $q=0$ including such fundamental issues as the presence or absence of algebraic
nonanalyticities does not contradict the RG irrelevance of the momentum dependence of $V(q)$ in the TLM.
From this, universality can only be deduced if {\it all} energy scales are sent to zero (see above).
A fixed $k - k_{\rm F} \neq 0$ sets a scale which becomes active for all generic $V(q)$
with $V^{(n)}(q=0) \neq 0$ for some $n \in {\mathbb N}$, where $V^{(n)}(q)$ denotes the $n$-th derivative.
Thus $\rho^\gtrless(k,\omega)$ cannot expected to be universal on general grounds. In fact, a
finite scale will destroy the scale invariance of the model and thus `quantum critical' power-law
scaling. In the case of the at $q=0$ `infinitely flat' box potential this mechanism is not active as
long as $|k-k_{\rm F}| < q_{\rm c}$.
We here supplement Ref.~\onlinecite{Meden99} and provide very strong evidence that the spectral function
of the TLM at fixed $k - k_{\rm F} \neq 0$ and for a generic potential is not characterized
by power laws. The latter are only found if $k - k_{\rm F} \to 0$.
To guide the reader we should from the outset be very precise about our understanding of
`power-law scaling'. We say that some real function $f$ shows power-law scaling with exponent $\xi$
close to $x_0 \in {\mathbb R}$ from above if $d \ln |f(x)| / d \ln(x-x_0)$ approaches $\xi$ for
$x \to x_0^+$. A similar definition can be given for power-law scaling from below.
This does of course not exclude that $f$ can to some degree be approximated by a power law or
`resembles' a power law for some range of $x$ (close to $x_0$) even if it does not fulfill
the above criterion. We will return to this in Sects.~\ref{sect_g4} and
\ref{sect_g2_g4} in which we present our results for $\rho^\gtrless(k,\omega)$.
More generally, we show that even for $|k-k_{\rm F}| \ll q_{\rm c}$ the spectral line
shape depends on the details of $V(q)$.
Although the present study might be regarded as somewhat technical -- or
even pedantic given that issues of the momentum dependence of $V(q)$ in the TLM are virtually
always nonchalantly ignored -- our results have far reaching consequences.
\subsection{Implications of our results}
The first implication of our results concerns the concept of LL universality. While $\rho^\gtrless(\omega)$ of
{\it any} model from the LL universality class shows the power-law suppression of spectral weight for
$\omega \to 0$ and $\rho^\gtrless(k_{\rm F},\omega)$ a power-law peak or suppression at the same energy,
{\it LL universality does not predict power laws in} $\rho^\gtrless(k,\omega)$ for fixed $k - k_{\rm F} \neq 0$.
Evidently, if this type of universality cannot be proven in the low-energy fixed point model, the TLM,
it cannot be a characteristic feature of the LL universality class.
This does of course not exclude that {\it certain models} from the LL universality class might show such
power laws, however, if so for {\it more specific reasons than LL universality.} An obvious example
for this is the TLM with box potential.\cite{Schoenhammer93} Other examples might be based upon the
restriction of the (equilibrium) dynamics encountered in specific 1d models with an extensive
number of local conserved quantities (e.g. the Hubbard model) which are often Bethe ansatz
solvable.\cite{Essler05} A detailed discussion in
which we relate our results to spectra of 1d lattice models obtained by either analytical or numerical
approaches is given in Sect.~\ref{sect_con_LLun}.
Strongly linked to this are the implications
of our findings for recent results on $\rho^\gtrless(k,\omega)$ taking the
nonlinearity of the single-particle dispersion in to
account\cite{Khodas07,Pereira09,Imambekov09a,Imambekov09b,Schmidt10,Essler15} which are
embedded in the framework of the so-called
`nonlinear LL phenomenology'.\cite{Imambekov12} In
this power laws are not viewed as originating from quantum critical scale invariance but
rather as resulting from a Fermi edge singularity like effect. In the phenomenological
construction of the effective field theory including curvature effects of the dispersion the above
described ad hoc regularization is employed. The spectral function is computed for this field theory.
Our results obtained for {\it linear} LL theory raise the question whether the power laws
found in `nonlinear LL phenomenology' are robust against a curvature of the bulk two-particle
potential. We emphasize that these
power laws are specific to the nonlinearity of the dispersion, which e.g. leads to momentum
dependent exponents, and are thus different from the nonanalyticities found for the TLM treated
within the ad hoc procedure (or, for that matter, the TLM with box potential). Again this does
of course not exclude that for {\it specific 1d models,} e.g. Bethe ansatz solvable lattice
models, power laws with momentum dependent exponents might be realized. More on this is can be found
in Sect.~\ref{sect_con_nonlinLL}.
Finally, our findings are of importance for the interpretation of experimental momentum resolved
spectra. Even after decades of research none of the photoemission experiments on quasi 1d
metals reporting on the observation of dispersing spin and charge peaks remains unchallenged.\cite{Grioni09}
One reason for this is that, when interpreting experimental data in the light of LL physics,
the momentum resolved spectral function obtained within the TLM employing an ad hoc
regularization is taken paradigmatically. Crucially, we find spin and charge peaks
for generic $V(q)$ and a sufficiently small amplitude of the two-particle interaction
even though they are not given by power-law singularities. Our results show, however, that details
are model dependent
(in our case $V(q)$ dependent) and therefore nonuniversal. Thus the detailed spectral features of the
ad hoc regularized TLM cannot expected to be found in experimental spectra. Further account of the
relevance of our results for experimental spectra is given in Sect.~\ref{sect_con_exp}.
\subsection{Structure of the paper}
The rest of our paper is structured as follows. In Sect.~\ref{sect_TLM} we introduce the TLM
and its bosonization solution. Constructive bosonization of the field operator can be used
to derive a closed analytical expression for the single-particle Green function, which, however,
contains a momentum integral on the right hand
side. This is discussed in Sect.~\ref{sect_Green}. In Sect.~\ref{sect_adhoc} we introduce
different versions of the ad hoc regularization to obtain integral-free expressions
for $G^\gtrless(x,t)$. Sections \ref{sect_box} and \ref{sect_arbitrary} are devoted
to the technical details of how we obtain exact spectra for a box potential and arbitrary
potentials, respectively.
In Sect.~\ref{sect_g4} we present results for the spectral function
of the so-called $g_4$-model with intra-branch interaction only and different shapes
of momentum dependency of the interaction. These are compared to
the ones obtained by the ad hoc procedure. Section~\ref{sect_g2_g4} ist devoted
to the spectral function of the spinless TLM -- the spinless
$g_2$-$g_4$-model. In Sect.~\ref{sect_discussion} we
discuss the implications of our results. When alluding to the relevance
of our insights for the interpretation of photoemission data we in addition
present spectral functions for the spinful TLM.
\section{The Tomonaga-Luttinger model}
\label{sect_TLM}
We here do not introduce the TLM by `deriving' it from the 1d interacting electron gas under certain
assumptions (e.g. on the real-space range of the interaction; no $2k_{\rm F}$ two-particle
scattering processes)\cite{Schoenhammer05} or as the effective field theory for microscopic
lattice models\cite{Giamarchi03} but rather take it as a stand-alone model.
It consists of independent right- ($\alpha=+$) and left-moving ($\alpha=-$) fermions with spin $s$, creation
operators $a^{\dag}_{k,\alpha,s}$, dispersion $\xi_{\alpha}(k)=
\alpha v_{\rm F} (k- \alpha k_F)$,
density operators ($q \neq 0$) $\rho_{\alpha,s}(q)= \sum_k
a_{k,\alpha,s}^{\dagger} a_{k+q,\alpha,s}^{}$, and particle
number operators $n_{k,\alpha,s}=a_{k,\alpha,s}^{\dagger}
a_{k,\alpha,s}^{}$. Following Luttinger\cite{Luttinger63} an infinite `Dirac sea' filled in the ground state
is assumed and thus the momentum quantum number $k$ of both particle species is unbounded. This simplifies
the mathematical treatment as certain relations become operator identities\cite{Haldane81,Schoenhammer05}
and are not only restricted to the low-energy part of the Hilbert space as in Tomonagas
approach.\cite{Tomonaga50} This addition of states often requires normal ordering.
The Hamiltonian for a system of length $L$ is given by
\begin{eqnarray}
&& H = \sum_k \sum_{\alpha,s} \xi_{\alpha}(k)
\left[ n_{k,\alpha,s} - \left< n_{k,\alpha,s} \right>_0 \right] \nonumber \\
& & + \frac{1}{2 L} \sum_{{q \neq 0} \atop {\alpha, s, s'}}
\left[g_{4,\parallel}(q) \delta_{s,s'} +
g_{4,\perp}(q) \delta_{s,-s'} \right]
\rho_{\alpha,s}(q) \rho^{\dag}_{\alpha,s'}(q) \nonumber \\
&& + \frac{1}{L} \sum_{{q \neq 0} \atop {s, s'}}
\left[g_{2,\parallel}(q) \delta_{s,s'} +
g_{2,\perp}(q) \delta_{s,-s'} \right]
\rho_{+,s}(q) \rho^{\dag}_{-,s'}(q) .
\label{hamiltonian}
\end{eqnarray}
Here $\left< \ldots \right>_0$ denotes the
(noninteracting) ground state expectation value (normal ordering).
We keep the explicit $q$-dependence of the two-particle potential.
The inter- ($g_2$) and intra-branch ($g_4$) potentials are not
necessarily equal and replace the potential $V(q)$ referred to in the Introduction.
Similarly, the interaction of spin parallel ($\parallel$) and anti-parallel
($\perp$; this is a confusing but standard notation\cite{Solyom79}) particles is not necessarily the same.
If the TLM is considered as the low-energy fixed point model of the LL universality class
this flexibility is required. The low-energy physics
of any model from this class is characterized by four independent numbers, e.g. the
two LL parameters $K_{c/s}$ and the two velocities
$v_{c/s}$.\cite{Haldane81,Schoenhammer05} In the TLM for a given $v_{\rm F}$ (and $k_{\rm F}$)
those are fixed by the (in general) four independent `coupling constants'
$g_{i,\kappa}(q=0)$ ($i=2,4$; $\kappa=\parallel , \perp$) at
vanishing momentum transfer (see below). Therefore, to encounter nontrivial interaction effects
in the LL sense the $g_{i,\kappa}(q=0)$ should not all be 0. We
restrict ourselves to these kind of interactions.
We assume that the Fourier transforms $g_{i,\kappa}(q)$ of the two-particle
interaction are even and for $q \geq 0$ monotonic functions which vanish for
$|q| \gg q_{\rm c}$, with an interaction cutoff $q_{\rm c}$.
These requirements are physically sensible if the TLM is considered in its own right.\cite{Tomonaga50}
If, however, the TLM is studied as the effective low energy model it is less clear
if this assumption holds. It is thus crucial that it the assumption is not essential
for our main conclusions. Relaxing it would merely complicate the calculations as
positive and negative momenta would have to be treated seperately.
We emphasize that at {\it no} stage of the discussion it will be necessary to introduce any
further ultraviolet cutoffs `by hand' despite the infinite (filled) Dirac sea at negative energies.
In this sense the Hamiltonian Eq.~(\ref{hamiltonian}) represents a mathematically well defined model.
To be more precise the Hamiltonian in Eq.~(\ref{hamiltonian}) defines a {\it whole class of models}
as the four coupling functions $g_{i,\kappa}(q)$ can be arbitrarily chosen as long as the
introduced requirements are fulfilled. Still, we continue to refer to this class as {\it the} TLM.
Note that a Hamiltonian of the form Eq.~(\ref{hamiltonian}) but with {\it coupling constants}
$g_{i,\kappa}$ instead of {\it coupling functions} $g_{i,\kappa}(q)$ can often be found in the literature.
The necessary ultraviolet regularization is then left implicit and frequently not uniquely defined.
We note that particle number contributions to the Hamiltonian which appear if the model is derived
from the interacting 1d electron gas\cite{Haldane81,Schoenhammer05} are suppressed as they do not
play any role for our considerations.
The spinless
version of the TLM follows from Eq.~(\ref{hamiltonian}) by dropping the spin index and keeping only
$g_{i}(q)$ instead of $g_{i,\kappa}(q)$ for $i=2,4$.
Bosonization of the Hamiltonian and a canonical transformation
lead to\cite{Haldane81,Schoenhammer05}
\begin{eqnarray}
H = \sum_{q \neq 0} \sum_{\nu =c,s} \omega_{\nu}(q)
\beta_{\nu}^{\dagger}(q) \beta_{\nu}^{}(q) + E_0,
\label{hamiltoniandiag}
\end{eqnarray}
with bosonic operators $\beta_{\nu}^{(\dag)}(q)$ describing collective
charge ($\nu=c$) and spin ($\nu=s$) excitations
(spin-charge separation) as well as the ground state energy $E_0$.
The $\beta_{\nu}^{(\dag)}(q)$ are linearly related to the densities $\rho_{\alpha,s}(q)$
of the fermions. The mode energies $\omega_{\nu}(q)$ are given by
\begin{eqnarray}
\frac{\omega_{\nu}(q)}{|q|} = v_{\rm F} \sqrt{\left( 1+
\frac{g_{4,\nu}(q)}{\pi v_{\rm F}} \right)^{\!\! 2} -
\left( \frac{g_{2,\nu}(q)}{\pi v_{\rm F}}\right)^{\!\! 2} }
= v_{\nu}(q) ,
\label{energies}
\end{eqnarray}
where we have introduced the renormalized {\it momentum-dependent}
charge and spin density velocities $v_{\nu}(q)$ and interactions
\begin{eqnarray}
g_{i,c/s}(q) = \left[g_{i,\parallel}(q)
\pm g_{i,\perp}(q) \right]/2.
\end{eqnarray}
We already now emphasize that for momentum dependent $g_{i,\kappa}(q)$ the eigenmode
dispersions will become {\it nonlinear}. The {\it linearization} of the latter,
that is the replacement
\begin{eqnarray}
v_{\nu}(q) \to v_{\nu}(0) = v_{\nu},
\label{vcs}
\end{eqnarray}
is the {\it crucial step} in the ad hoc regularization procedure to derive
integral-free expressions for correlation functions (see below).
%However, it is
%also the step which leads to the power laws (as a function of $\omega$)
%in $\rho^\gtrless(k,\omega)$ for fixed $k - k_{\rm F} \neq 0$; any curvature
%of $v_{\nu}(q)$ at $q=0$, that is any curvature of $g_{i,\kappa}(q)$,
%destroys such power laws.
In Eq.~(\ref{vcs}) we have introduced the $q$-independent charge and spin velocities
$v_{c/s}$ relevant for the low-energy physics (all energy scales sent to 0) in the LL sense. The
corresponding LL parameters $K_{c/s}$ of the TLM are obtained from
\begin{eqnarray}
K_{\nu} (q)= \sqrt{\frac{1+g_{4,\nu}(q)/(\pi v_F)-g_{2,\nu}(q)/(\pi
v_F)}{1+g_{4,\nu}(q)/(\pi v_F)+g_{2,\nu}(q)/(\pi
v_F)}}
\label{Kdef}
\end{eqnarray}
in the limit $q \to 0$. Note that $K_{\nu} (q) = 1$ if the inter-branch interaction
$g_{2,\nu}(q)$ vanishes.
The spinless version of the bosonized Hamiltonian is obtained after dropping
the terms with index $\nu=s$.
When presenting our results for $\rho^\gtrless(k,\omega)$ we will initially focus on
two special cases. The first is
the spinful TLM with intra-branch interaction only, that is $g_{2,\kappa}(q)=0$, which is
commonly referred to as the $g_4$-model. The second is the spinless $g_2$-$g_4$-model.
They are paradigmatic for the two interaction effects characteristic for LLs: spin-charge
separation and power-law scaling with interaction dependent exponents, respectively.
Proceeding this way increases the transparency of our analysis.
The scenario for the general spinful model can be deduced by combining the insights of
both cases; in Sect.~\ref{sect_con_exp} we in addition present a few explicit results
for the spinful TLM.
\section{The momentum resolved spectral function}
\subsection{The single-particle Green function}
\label{sect_Green}
We are interested in the momentum resolved spectral function.
It can be computed by Fourier transforms from the greater and lesser
single-particle Green function
\begin{align}
i G^>_{\alpha,s} (x,t) & = \langle \psi_{\alpha,s} (x,t) \psi_{\alpha,s}^\dagger (0,0) \rangle \label{Green} \\
i G^<_{\alpha,s} (x,t) & = \langle \psi_{\alpha,s}^\dagger (0,0) \psi_{\alpha,s} (x,t) \rangle.
\end{align}
The field operators $\psi_{\alpha,s}^\dagger(x)$ and $\psi_{\alpha,s}(x)$ are related in the usual way
to the creation and annihilation operators in momentum space
\begin{equation}
\psi_{\alpha,s} (x)= \frac{1}{\sqrt{L}} \sum_k e^{i k x} a_{k,\alpha,s}.
\end{equation}
Particle-hole symmetry of the TLM ensures\cite{Meden92,Schoenhammer05}
\begin{equation}
G^>_{\alpha,s} (x,t) = G^<_{\alpha,s} (-x, -t).
\end{equation}
and it is sufficient to consider the greater Green function.
We note in passing that band filling is not an issue in the TLM as
we consider it. However, when the model is investigated as the effective
low-energy model of another 1d correlated electron model the band filling
of the latter will enter in the LL parameters and
velocities characterizing the low-energy physics.\cite{Schoenhammer05}
Furthermore, with $x \rightarrow -x$ we can go over from the Green function of
right-movers to the one of left-movers. Therefore, we will only study
$G^>_{+,s}(x,t)$. In the absence of a magnetic field the Green function is spin independent and
we thus suppress the spin index from now on.
To compute the ground state expectation value in Eq.~(\ref{Green}) we use constructive bosonization
of the field operator.\cite{Haldane81,Schoenhammer05} We emphasize that this approach is based on
operator identities and does not require the introduction of any cutoffs if one first considers
finite systems of length $L$ (as we will do). In more phenomenological approaches\cite{Giamarchi03} a cutoff
is introduced -- often denoted by $1/\alpha$ and referred to as an `effective band width' -- which
formally has to be send to infinity. However, it is frequently kept finite artificially. This is
part of one of the possible ad hoc ultraviolet regularizations.
The exact greater Green function of right-movers for the most general spinful $g_2$-$g_4$-model is given by
\begin{equation}
i \,G^>_{+} (x,t) =i \, \left[ G^>_{+} \right]^0 (x,t) \, e^{F(x,t)},
\label{eq:Ggtrplus_spin}
\end{equation}
with
\begin{equation}
\left[ G^>_{+} \right]^0 (x,t) = \frac{ e^{i (k_{\rm F}+\pi/L) x}}{L}
\exp \left\{ \sum_{n=1}^\infty \frac{1}{n} e^{ i q_n (x- v_{\rm F} t + i 0^+)} \right\}.
\label{freeGreen}
\end{equation}
The interaction enters in
\begin{align}
F(x,t) = & \frac{1}{2} \sum_{\nu = c,s} \sum_{n=1}^\infty \frac{1}{n}
\left[ e^{ i q_n x } \left( e^{- i \omega_\nu(q_n) t} - e^{ - i v_{\rm F} q_n t} \right) \right.
\nonumber \\
& + \left. 2 \gamma_\nu(q_n)
\left( \cos (q_n x) e^{- i \omega_\nu(q_n) t} -1 \right) \right],
\label{eq:F_g2g4spinful}
\end{align}
with
\begin{eqnarray}
\gamma_{\nu}(q)=\left[ K_{\nu}(q)+ 1/K_{\nu}(q) -2 \right]/4
\label{gammadef}
\end{eqnarray}
and $q_n=n 2\pi/L $ (periodic boundary conditions).
Due to the decay of the $g_{i,\kappa}(q)$ on the scale $q_{\rm c}$ the momentum sum in
$F(x,t)$ is convergent in the ultraviolet.
Note in particular that for $q_n \gg q_{\rm c}$ the two terms in the first line cancel each other as
$\omega_\nu(q_n) \to v_{\rm F} q_n$ in this limit [compare Eq.~(\ref{energies})].
The term in the second line of Eq.~(\ref{eq:F_g2g4spinful}) is convergent as
$\gamma _\nu(q_n) \to 0$ for $q_n \gg q_{\rm c}$ [compare Eqs.~(\ref{gammadef}) and (\ref{Kdef})].
For vanishing interaction $F(x,t)=0$. Thus $\left[ G^>_{+} \right]^0 (x,t)$ is the noninteracting Green function.
In the thermodynamic limit $L \to \infty$ it becomes
\begin{equation}
\left[ G^>_{+} \right]^0 (x,t) = \frac{1}{2 \pi} \frac{e^{i k_{\rm F} x}}{x-v_{\rm F}t +i 0^+} .
\label{exactGreen}
\end{equation}
The factor $\exp{(- q_n 0^+)}$ in Eq.~(\ref{freeGreen}) ensuring convergence appears naturally, and is
not related to any ad hoc regularization. Only with this factor the exactly known $ \left[ G^>_{+} \right]^0 (x,t)$
Eq.~(\ref{exactGreen}) and from this the exact noninteracting spectral function
\begin{eqnarray}
\left[ \rho^>_{+} (k, \omega ) \right]^0 (x,t) = \Theta(k-k_{\rm F}) \delta(\omega-\xi_{+}(k))
\end{eqnarray}
can be obtained. Due to the linear single-particle dispersion $ \left[ G^>_{+} \right]^0 (x,t)$
is of relativistic form.
The greater spectral function is defined as
\begin{equation}
\rho^>_{+} (k_n, \omega ) = \frac{1}{2 \pi} \int_{-\infty}^\infty
dt \, e^{i \omega t} \int_{-L/2}^{L/2} dx \, e^{- i k_n x} \; i G^>_{+} (x,t) .
\end{equation}
As $G^>_{+} (x,t)$ Eqs.~(\ref{eq:Ggtrplus_spin}) to (\ref{eq:F_g2g4spinful}) is an analytic function
in the lower half of the complex $t$-plane $\rho^>_{+} (k, \omega ) $
has nonvanishing weight only for $\omega \geq 0$.
To compute $\rho^>_{+} (k, \omega )$ for arbitrary potentials in the thermodynamic limit
three nested integrals have to be performed. The integrands are slowly decaying,
oscillatory, and have poles close to the real axis. Therefore, the accuracy which can
be achieved when straightforwardly performing the integrals numerically, given $g_{i,\kappa}(q)$, is
not sufficient to answer the question whether or not power laws can be found for
$k-k_{\rm F} \neq 0$. We are thus forced to proceed differently. Before presenting
our approach to the exact spectral function of the TLM in Sects.~\ref{sect_box} and
\ref{sect_arbitrary} we will describe approximate ad hoc procedures which were pursued in
the literature to make analytical progress.
\subsection{The ad hoc regularization}
\label{sect_adhoc}
In the natural way of writing the {\it exact} Green function of
the TLM Eq.~(\ref{eq:Ggtrplus_spin}) the noninteracting one is factorized out.\cite{Meden99} However,
other expressions
for $G^>_{+} (x,t)$ can be found in the literature.\cite{Luther74,Meden92,Voit93,Voit95} These
result from the following procedure. The term in the curly brackets of Eq.~(\ref{freeGreen})
is canceled against the last term in the first line of Eq.~(\ref{eq:F_g2g4spinful}).
After this step the remaining $q$-sum in Eq.~(\ref{eq:F_g2g4spinful}) is no longer
convergent in the ultraviolet. Two ways have been reported how to deal with this problem.
In the first (i) the remaining first term of the first line of Eq.~(\ref{eq:F_g2g4spinful}) is
multiplied by $\exp{(- q_n 0^+)}$. To obtain integral-free expressions for $G^>_{+} (x,t)$ one then
assumes that $\gamma_{\nu}(q_n)$ is given by $\gamma_{\nu} \exp{(- q_n \Lambda)}$ which can be reached
by choosing a proper momentum dependence of the $g_{i,\kappa}(q)$. Thus one simply
selects a certain interaction potential. In {\it addition,} one linearizes the
eigenmode dispersions $\omega_{\nu}(q) \to v_\nu q$ for all $q$. This is {\it not}
a consequence of the special choice of the interaction potential but is rather an approximation
done independently `by hand'.
In the thermodynamic limit one then obtains
\begin{align}
[G_{+}^{>}]_{\rm (i)} (x,t) = & \frac{e^{i k_{\rm F} x}}{2 \pi}\prod_{\nu =c,s}
\left[ \frac{1}{x- v_{\nu} t +i 0^+ } \right]^{1/2} \nonumber \\*
& \times \left[ \frac{\Lambda^2}
{\left( x- v_{\nu} t + i \Lambda \right)
\left( x+ v_{\nu} t - i \Lambda \right) }
\right]^{\gamma_{\nu}/2} .
\label{gapprox1}
\end{align}
In the second ad hoc procedure (ii) the entire (remaining)
argument of the sum in Eq.~(\ref{eq:F_g2g4spinful}) is multiplied by $\exp{(- q_n \Lambda)}$ with
a {\it finite} momentum cutoff $1/\Lambda>0$. The momentum dependence of $\gamma_{\nu}(q_n)$ is dropped
and $\omega_{\nu}(q)$ is again linearized.
This leads to
\begin{align}
[G_{+}^{>}]_{\rm (ii)} (x,t) = & \frac{e^{i k_{\rm F} x}}{2 \pi}\prod_{\nu =c,s}
\left[ \frac{1}{x- v_{\nu} t +i \Lambda } \right]^{1/2} \nonumber \\*
& \times \left[ \frac{\Lambda^2}
{\left( x- v_{\nu} t + i \Lambda \right)
\left( x+ v_{\nu} t - i \Lambda \right) }
\right]^{\gamma_{\nu}/2} .
\label{gapprox2}
\end{align}
Finally, a third way (iii) to obtain an integral-free expression for $G^>_{+} (x,t)$ is based directly
on Eqs.~(\ref{eq:Ggtrplus_spin}) to (\ref{eq:F_g2g4spinful}). In this the
momentum dependence of $\gamma_{\nu}(q_n)$ is dropped and $\omega_{\nu}(q)$ is linearized after the
square bracket in Eq.~(\ref{eq:F_g2g4spinful}) was multiplied by $ \exp{(- q_n \Lambda)}$.
This way one obtains
\begin{align}
[G_{+}^{>}]_{\rm (iii)} (x,t) = & \frac{e^{i k_{\rm F} x}}{2 \pi} \frac{1}{x- v_{\rm F} t +i0^+} \nonumber \\*
& \times \prod_{\nu =c,s}
\left[ \frac{x- v_{\rm F} t + i \Lambda}
{x- v_{\nu} t + i \Lambda } \right]^{1/2} \nonumber \\*
& \times
\left[ \frac{\Lambda^2}
{\left( x- v_{\nu} t + i \Lambda \right)
\left( x+ v_{\nu} t - i \Lambda \right) }
\right]^{\gamma_{\nu}/2} .
\label{gapprox3}
\end{align}
We note that for the special case $\gamma_s=0$ (spin-rotational invariant interaction) and
$v_s = v_{\rm F}$ this is exactly the expression derived by Dzyaloshinski\v{i} and
Larkin \cite{Dzyaloshinskii73} in a purely fermionic approach which is based on Ward identities
and the closed loop theorem. \cite{Bohr81}
Obviously, all three {\it approximate} functions differ. However, their asymptotic behavior
(power-law decay) for large space-time arguments is the same, in particular along the four
special directions $x = \pm v_\nu t$. For this reason they all lead to the same
power laws in $\rho^>_{+} (k, \omega )$ for $\omega$ close to $\pm v_{\nu} (k-k_{\rm F})$
as derived in Refs.~\onlinecite{Meden92,Voit93,Voit95,Meden99}.
Similarly, for $x=0$ all three approximate expressions
have the same asymptotic behavior at large $|t|$, $\sim t^{-\alpha-1}$,
with $\alpha = \gamma_{c} + \gamma_{s}$ (and are analytic functions in the
lower half of the complex $t$-plane). After a single Fourier transform from time
to frequency this gives $\rho^{>}(\omega) \sim \Theta(\omega) \omega^\alpha$.
As discussed in Ref.~\onlinecite{Meden99} this power-law scaling of the momentum integrated spectral
function is also found based on the {\it exact} expressions (\ref{eq:Ggtrplus_spin}) to
(\ref{eq:F_g2g4spinful}) and for arbitrary $g_{i,\kappa}(q)$ (fulfilling the above mentioned restrictions)
with $\gamma_{\nu} \to \gamma_{\nu}(q=0)$. This universality is based on
the RG irrelevance of the momentum dependence of the interaction;\cite{Solyom79}
the appearance of the power law with an exponent set by the two-particle potential
at vanishing momentum transfer $q=0$ is not affected by the potential away from this point.
Any of the discussed ad hoc procedures (i) to (iii) can thus be employed without spoiling
the universal behavior. The same holds at $t=0$ but $x \neq 0$ which after Fourier transform
leads to the momentum distribution function $n_+(k)$ which also shows universal power-law scaling
for $k \to k_{\rm F}$.\cite{Voit95,Giamarchi03,Schoenhammer05}
The question we address here is whether or not the ad hoc regularized Green functions
can also be used to obtain universal results for $\rho^>_{+} (k, \omega )$ at $k-k_{\rm F} \neq 0$.
As already emphasized the linearization of the $\omega_{\nu}(q)$ is the crucial step which leads
to power laws in $\rho^>_{+} (k, \omega )$ for $k - k_{\rm F} \neq 0$ in the ad hoc procedures.
In Ref.~\onlinecite{Meden99} it was shown using a generalization of the stationary phase method that
any $q=0$-curvature of $v_\nu(q)$ affects the asymptotic behavior of $G^>_{+} (x,t)$ in the
distinguished directions $x = \pm v_\nu t$.
The decay is no longer solely given by the $\gamma_\nu$ and therefore not only by the $g_{i,\kappa}(0)$.
However, in the ad hoc procedures (with constant $v_\nu$), one can analytically
show that it is the asymptotic behavior in these directions of the $x$-$t$-plane which leads to the
power laws at $\omega = \pm v_\nu [k-k_{\rm F}]$.\cite{Meden99} This raises doubts that the latter
are generic.
Besides the spinful TLM we also consider its spinless version. From the above expressions
for $ G_{+}^{>} (x,t)$ [including the ones of the ad hoc regularization (i) to (iii)] the
spinless Green function is obtained by setting $\gamma_c=\gamma_s$,
$\omega_c=\omega_s$, and $v_c=v_s$. All what was said in the last two paragraphs
about power-law behavior and universality remains valid in the spinless case up to
the (obvious) exception that in the ad hoc regularized spectral function only the
two (instead of four) distinguished energies $\pm v_{c} (k-k_{\rm F})$ exist.
We note in passing that it was realized decades ago that within the
ad hoc procedures (i) and (ii) exact spectral sum rules are not
fulfilled.\cite{Suzumura80,Schoenhammer93b,Voit93b}
\subsection{How to compute $\rho^>_{+} (k, \omega )$ for a box potential}
\label{sect_box}
In the Introduction we mentioned that exact results for the momentum resolved spectral function
of the TLM were derived based on Eqs.~(\ref{eq:Ggtrplus_spin}) to
(\ref{eq:F_g2g4spinful}) assuming a box potential.\cite{Schoenhammer93} We will compare our results
for other potentials to these. To be self-contained we here give all the formulas required to obtain
$\rho^>_{+} (k, \omega )$ of the spinful $g_4$-model and the spinless $g_2$-$g_4$-model
for $g_{i,\kappa}(q)= g_{i,\kappa} \Theta(q_{\rm c}^2 - q^2)$. More details are presented for the first
case which was not separately discussed in Ref.~\onlinecite{Schoenhammer93}. We note
that for a reader primarily interested in results it is not necessary to understand the
technical details of this section in full detail.
\subsubsection{The spinful $g_4$-model}
\label{sping4box}
For $g_{2,\kappa}(q)=0$ it directly follows that $\gamma_\nu(q)=0$ as $K_\nu(q)=1$;
compare Eqs.~(\ref{gammadef}) and (\ref{Kdef}). Equations
(\ref{eq:Ggtrplus_spin}) to (\ref{eq:F_g2g4spinful}) for the exact Green function
thus simplify considerably. The same holds for the ad hoc regularized
expressions Eqs.~(\ref{gapprox1}) to (\ref{gapprox3}) as $\gamma_\nu =0$.
These are characterized by square-root singularities at $x=v_c t$ and
$x=v_s t$. After Fourier transform they lead to square root singularities in
$\rho^>_{+} (k, \omega )$ for
$\omega \to v_c [k-k_{\rm F}]$ and $\omega \to v_s [k-k_{\rm F}]$.\cite{Meden92,Voit93,Voit95}
We note that $[G_{+}^{>}]_{\rm (iii)} (x,t)$
contains the additional factor $(x- v_{\rm F} t + i \Lambda)/(x- v_{\rm F} t +i0^+)$
which does not drop out as $\Lambda>0$. Further down we will discuss how this term affects
the spectral properties. As the ground state of the $g_4$-model remains
the noninteracting one,\cite{Schoenhammer93}
$\rho^>_{+} (k, \omega )$ has finite weight only for $k \geq k_{\rm F}$.
For the case of a box potential, it is
\begin{equation}
v_\nu(q) =
\begin{cases}
v_\nu, & q \leq q_{\rm c} \\
v_{\rm F}, & q > q_{\rm c} .
\end{cases}
\end{equation}
Setting for convenience
$z = \exp \{ i (2 \pi / L ) ( x - v_{\rm F} t) \}$ as well as
$z_\nu = \exp \{ i (2 \pi / L ) ( x - v_\nu t) \}$ and expanding the exponential function one obtains
from Eqs.~(\ref{eq:Ggtrplus_spin}) to (\ref{eq:F_g2g4spinful})
\begin{align}
i G^>_{+} (x,t) & = \frac{1}{L} \,
e^{i (2 \pi/L) (n_{\rm F}+1) x } \: \left( \sum_{l=0}^\infty z^l \right) \notag \\
& \times \prod_{n=1}^{n_{\rm c}} \left( \sum_{m=0}^\infty \frac{ (-1/n)^m }{m!} z^{n m} \right)
\left( \sum_{j=0}^\infty \frac{ (1/(2n))^j }{j!} z_c^{n j} \right) \notag \\
& \times \left( \sum_{l=0}^\infty \frac{ (1/(2n))^l }{l!}
z_s^{n l} \right) \label{eq:g4spinG} \\
& \overset{!}{=} \frac{1}{L} \, e^{i (2 \pi/L) (n_{\rm F}+1) x } \: \left( \sum_{m=0}^\infty a_m^{(n_{\rm c})} z^m \right)
\notag \\
& \times
\left( \sum_{j=0}^\infty b_j^{(n_{\rm c})} z_c^j \right) \left( \sum_{l=0}^\infty b_l^{(n_{\rm c})}
z_s^l \right), \label{eq:g4spinG_coeff}
\end{align}
with $q_{\rm c}= n_{\rm c} 2\pi/L$.
Here $k_{\rm F} = (2 n_{\rm F} + 1) \pi /L$ and $n_{\rm F}$ is the index of the last occupied
single-particle state; $k_{\rm F}$ thus lies in between the last occupied and the first unoccupied
one.
The coefficients in Eq.~\eqref{eq:g4spinG_coeff} can be determined by a recursion relation,
where $m > 1$, $l \in \mathbb{N}_0$ and $i = 0, \dots, m-1$:
\begin{align}
& a_{lm+i}^{(m)} = \sum_{j=0}^l \frac{ (-1/m)^j}{j!} \ a_{m(l-j)+i}^{(m-1)} , \label{eq:a_rec} \\
& b_{lm+i}^{(m)} = \sum_{j=0}^l \frac{ (1/(2m))^j}{j!} \ b_{m(l-j)+i}^{(m-1)} .
\end{align}
The initial values are given by
\begin{equation}
a_m^{(1)} = \sum_{j=0}^m \frac{ (-1)^j }{ j! } \label{eq:a_rec_ini}
\end{equation}
and $b_m^{(1)} = (1/2)^m/m!$. The recursion can easily be performed on a computer.
The double Fourier transform can be computed analytically and one obtains
\begin{align}
\rho^>_{+} (k_n, \omega) & = \sum_{l=0}^{\tilde n} \sum_{j=0}^{\tilde n-l}
a_{\tilde n-l-j}^{(n_{\rm c})} \, b_l^{(n_{\rm c})} \, b_j^{(n_{\rm c})} \notag \\
& \times
\delta \left[ \omega - \frac{2 \pi }{L} \big( (\tilde n-l-j) v_{\rm F} + l v_c + j v_s \big) \right] ,
\label{sping4rec}
\end{align}
with $\tilde n= n-(n_{\rm F} +1)$. This way the exact spectral function $\rho^>_{+} (k_n, \omega)$ of the TLM with
box potential can easily be computed for large but finite systems (see Sect.~\ref{sect_g4}).
To obtain analytical insights we rewrite the Green function as
\begin{align}
i G^>_{+} (x,t) & = \frac{1}{L} \, e^{i (2 \pi/L) (n_{\rm F}+1) x } \notag \\
& \times
\exp \left\{ \frac{1}{2} \sum_{n=1}^{n_{\rm c}} \left( \frac{ z_c^n}{n} + \frac{ z_s^n }{n} \right) +
\sum_{n=n_{\rm c}+1}^\infty \frac{z^n}{n} \right\}.
\end{align}
Thus
\begin{equation}
\sum_{m=0}^\infty a_m^{(n_{\rm c})} z^m = 1 + \sum_{n=n_{\rm c}+1}^\infty \frac{z^n}{n}
+ \frac{1}{2} \left( \sum_{n=n_{\rm c}+1}^\infty \frac{z^n}{n} \right)^2 + \ldots
\end{equation}
and we immediately see that $a_0^{(n_{\rm c})} = 1$, $a_m^{(n_{\rm c})} = 0$ for $1 \leq m \leq n_{\rm c}$ and $a_m^{(n_{\rm c})}
= 1/m$ for $ n_{\rm c}+1 \leq m \leq 2 n_{\rm c} +1$.
For $\tilde{n} \leq n_{\rm c}$ the simplified expression
\begin{equation}
\rho^>_{+} (k_n, \omega)
= \sum_{l=0}^{\tilde{n}} b_l^{(n_{\rm c})} b_{\tilde{n}-l}^{(n_{\rm c})} \ \delta \left[ \omega -
\frac{2 \pi }{L} \left( \tilde{n} v_s + l (v_c - v_s) \right) \right]
\label{eq:g4spinrhoex}
\end{equation}
holds. For fixed $\tilde{n}$ and $v_s < v_c$, there is only spectral weight for
$v_s [k-k_{\rm F}] \leq \omega \leq v_c [k-k_{\rm F}]$ (up to corrections of order $1/L$).
For $v_s > v_c$ the roles of the two velocities are interchanged.
We note that for $k-k_{\rm F} \leq q_{\rm c}$ the bare Fermi velocity $v_{\rm F}$
drops out.
Further analytical results can be obtained employing
\begin{align}
& \exp \left\{ \alpha \sum_{n=1}^{n_{\rm c}} \frac{1}{n} x^n \right\} = (1-x)^{-\alpha} \, \exp \left\{- \alpha \sum_{n=n_{\rm c}+1}^\infty \frac{1}{n} x^n \right\} \notag \\
& = \left[ \sum_{j=0}^\infty \binom{-\alpha}{j} (-x)^j \right] \, \exp \left\{- \alpha \sum_{n=n_{\rm c}+1}^\infty \frac{1}{n} x^n \right\}, \label{eq:Identity_expgammasum}
\end{align}
where $\binom{-\alpha}{j}$ is the generalized binomial coefficient.
From this it follows that $b_0^{(n_{\rm c})} = 1$ and
\begin{equation}
b_j^{(n_{\rm c})} = (-1)^j \binom{-1/2}{j} \quad \xrightarrow{1 \ll j \leq n_{\rm c}} \: \text{const.} \times j^{-1/2},
\end{equation}
where
\begin{equation}
(-1)^j \binom{-\alpha}{j}
\approx \frac{1}{ \Gamma[\alpha] } j^{\alpha-1} \quad \text{for } j \rightarrow \infty
\end{equation}
was used.
Inserting this for energies close to $ v_s [k_n-k_{\rm F}]$ into Eq.~\eqref{eq:g4spinrhoex}, where
$l$ is the integer next to $(\omega - v_s [k_n -k_{\rm F}] ) / [(2 \pi/L) (v_c - v_s)]$, we obtain for
$L \rightarrow \infty$ the one-sided square-root singularity ($v_c > v_s$)
\begin{equation}
\rho^>_{+} (k, \omega) \sim \Theta
\left( \omega- v_s [k-k_{\rm F}] \right)
\left( \omega - v_s [k-k_{\rm F}] \right)^{-1/2}.
\label{wurz}
\end{equation}
Analogously we get close to $ v_c [k_n-k_{\rm F}]$
\begin{equation}
\rho^>_{+} (k, \omega) \sim \Theta
\left( \omega- v_c [k-k_{\rm F}] \right)
\left(\omega - v_c [k-k_{\rm F}] \right)^{-1/2}.
\end{equation}
For $v_s > v_c$ we only need to interchange the two velocities.
We have thus shown that the exact spectral
function of the $g_4$-model with box potential shows the edge singularities also found within
the ad hoc procedures (i) to (iii).
We note in passing that for $k=k_{\rm F}$ the spectral function reduces to a $\delta$-function of
weight 1 located at $\omega=0$.
\subsubsection{The spinless $g_2$-$g_4$-model}
\label{nospinbox}
In Sect. IV of Ref.~\onlinecite{Schoenhammer93} a recursive way of computing the spectral function for the
full spinless TLM with box potential similar to Eqs.~(\ref{eq:a_rec})-(\ref{sping4rec}) was introduced.
It is given by
\begin{align}
\rho^>_+(k_n, \omega) & = A^{-2 \gamma_c}
\sum_{r=\max\{0, -\tilde{n}\}}^\infty \sum_{l=0}^{\tilde{n}+r} a_{\tilde{n}+r-l}^{(n_{\rm c})} b_l^{(n_{\rm c})}
c_r^{(n_{\rm c})} \notag \\
& \times \delta \left[ \omega - \frac{2 \pi}{L} \big( ( \tilde{n}+r-l ) v_{\rm F} +
(r+l) v_c \big) \right],
\label{eq:g2g4rho_boxex}
\end{align}
with $A = \exp\{ \sum_{n=1}^{n_{\rm c}} (1/n) \}$.
The coefficients $a_m^{(n_{\rm c})}$ are determined as in Eq.~\eqref{eq:a_rec} and \eqref{eq:a_rec_ini}.
For $b_m^{(n_{\rm c})}$, the recursion relation reads
\begin{equation}
b_{lm+i}^{(m)} = \sum_{j=0}^l \frac{1}{j!} \left( \frac{1+\gamma_c}{m} \right)^j \ b_{m(l-j)+i}^{(m-1)} ,
\end{equation}
with the initial values $b_m^{(1)} = (1+\gamma_c)^m/m!$, and for the $c_m^{(n_{\rm c})}$
\begin{equation}
c_{lm+i}^{(m)} = \sum_{j=0}^l \frac{1}{j!} \left( \frac{\gamma_c}{m} \right)^j \ c_{m(l-j)+i}^{(m-1)} ,
\end{equation}
with the initial values $c_m^{(1)} = \gamma_c^m/m!$.
In analogy to the considerations for the spinful $g_4$-model, we can infer from
the behavior of the coefficients the behavior of the spectral function close to
$v_c [k-k_{\rm F}]$ (for $L \to \infty$).
For $k-k_{\rm F} >0 $ we find\cite{Schoenhammer93}
\begin{equation}
\rho^>_+(k, \omega) \sim \Theta( \omega - v_c [k-k_{\rm F}] ) \
\left( \omega - v_ c[k-k_{\rm F}] \right)^{\gamma_c -1},
\label{power1}
\end{equation}
that is a divergence if $\gamma_c <1$ (not to strong interactions).
In contrast to the $g_4$-model $\rho^>_+$ can now also carry spectral weight
for $(k-k_{\rm F}) <0 $. At the threshold we obtain\cite{Schoenhammer93}
\begin{equation}
\rho^>_+(k, \omega) \sim \Theta( -\omega - v_c [k-k_{\rm F}] ) \
\left( -\omega - v_c [k-k_{\rm F}] \right)^{\gamma_c},
\label{power2}
\end{equation}
that is a power-law suppression since $\gamma_c > 0$ in the full model.
In the special case $k=k_{\rm F}$, power-law behavior with $\rho^>_+(k_{\rm F}, \omega)
\propto \omega^{2 \gamma_c -1}$ is found. These threshold power laws can also be found
based on the Green function of the ad hoc procedures (i) to (iii) discussed in
Sect.~\ref{sect_adhoc}.\cite{Theumann67,Luther74,Meden92,Voit93}
\subsection{How to compute $\rho^>_{+} (k, \omega )$ for arbitrary potentials}
\label{sect_arbitrary}
We now show that expressions for the Green and the spectral function which involve
recursively computed coefficients can also be given for an arbitrary
momentum dependence of the two-particle potential.
For the box potential the dispersion of the elementary charge and spin modes
is piecewise linear. This changes for arbitrary potentials. As a consequence
the coefficients become time-dependent and the Fourier transform with respect to time
has to be performed numerically, e.g. using fast Fourier transform (FFT).
\subsubsection{The spinful $g_4$-model}
\label{sping4arb}
For arbitrary potentials the spin and charge velocity are no longer piecewise
momentum independent. If the potential is effectively zero for $ q > \tilde{q}_{\rm c}$,
with a $\tilde{q}_{\rm c}$ which we take sufficiently larger then the characteristic scale
$q_{\rm c}$, we can work with $g_{4,\nu} (q) \rightarrow g_{4,\nu} (q) \, \Theta(\tilde q_{\rm c}^2 - q^2)$
for all practical purposes. Then we can rewrite $F(x,t)$ in Eq.~(\ref{eq:F_g2g4spinful})
\begin{align}
F(x,t) & = \sum_{n=1}^{\tilde{n}_{\rm c}} \frac{1}{n} \Bigg( \frac{1}{2}
e^{-i \omega_c(q_n) t} +\frac{1}{2} e^{-i \omega_s(q_n) t} \notag \\
& - e^{-i (2 \pi/L) n v_{\rm F} t} \Bigg) \, z^n ,
\end{align}
with $ z = \exp \{ i (2 \pi/L) x \}$, and use \textit{time-dependent} coefficients
to write
\begin{align}
i G^>_{+} (x,t) &= \frac{1}{L} \, e^{i (2 \pi/L) (n_{\rm F}+1) x } \:
\left( \sum_{l=0}^\infty z^l \, e^{i (2 \pi/L) l v_{\rm F} t } \right) \notag \\
& \times \prod_{n=1}^{\tilde{n}_{\rm c}} \Bigg( \sum_{m=0}^\infty \frac{(1/n)^m}{m!}
\Bigg[ \frac{1}{2} e^{- i \omega_c(q_n) t} \notag \\ & +\frac{1}{2} e^{-i \omega_s(q_n) t}-
e^{-i (2 \pi/L) n v_{\rm F} t} \Bigg]^m z^{m n}\Bigg) \\
& \overset{!}{=} \frac{1}{L} \, e^{ i (2 \pi/L) (n_{\rm F}+1) x } \sum_{m=0}^\infty a_m^{(\tilde{n}_{\rm c})} (t)
\, z^m.
\end{align}
The recursion relation for the $a_m^{(\tilde{n}_{\rm c})}(t)$ is given by
\begin{align}
& a_m^{(1)} (t) = \sum_{l=0}^m \frac{1}{l!} \left[ e^{-i(2 \pi/L) v_{\rm F} t } \right]^{m-l} \notag \\
& \times
\left[ \frac{1}{2} e^{-i \omega_c(q_1) t } + \frac{1}{2} e^{-i \omega_s(q_1) t } -
e^{-i(2 \pi/L) v_{\rm F} t } \right]^l , \\
& a_{lm+i}^{(m)} (t) = \sum_{j=0}^l \frac{(1/m)^j}{j!} \Bigg[ \frac{1}{2} e^{-i \omega_c(q_m) t } \notag \\ & +
\frac{1}{2} e^{-i \omega_s(q_m) t } - e^{-i(2 \pi/L) m v_{\rm F} t } \Bigg]^j \ a_{m(l-j)+i}^{(m-1)}(t),
\end{align}
where in the second line $m > 1$, $l \in \mathbb{N}_0$ and $i = 0, \dots, m-1$.
With this representation of the Green function, the Fourier transform to momentum space
can be performed analytically
\begin{equation}
i G^>_{+} (k_n,t) = \frac{1}{2 \pi} a_{\tilde{n}}^{(\tilde{n}_{\rm c})} (t).
\label{eq:g4spinG(k,t)}
\end{equation}
The remaining Fourier transform
\begin{equation}
\rho^>_{+} (k_n, \omega) = \int_{-\infty}^\infty d t \ e^{i \omega t} \ i G^>_+ (k_n, t)
\end{equation}
can be performed numerically as a FFT. Since for the finite system the spectral function
consists of $\delta$-peaks, the Green function does not decay in time. Therefore, we have to multiply
$i G^>_+(k_n,t)$ with a decaying function before performing the FFT. Here, we will always use the
exponential function $\exp \{ -\chi |t| \}$. In frequency space, this corresponds to a convolution of
the spectral function with the Lorentzian $\pi^{-1} \chi / ( \omega^2 + \chi^2) $, i.e. each $\delta$-peak
is broadened into a Lorentzian of width $\chi$.
\subsubsection{The spinless $g_2$-$g_4$-model}
\label{nospinarb}
In the same way as for the spinful $g_4$-model, we can introduce recursively defined time-dependent
coefficients for the spinless $g_2$-$g_4$-model
\begin{align}
a_m^{(1)} (t) & = \sum_{l=0}^m \frac{1}{l!} \left[ e^{-i (2 \pi/L) v_{\rm F} t} \right]^{m-l} \notag \\
& \times \left[ \left( 1+\gamma_c(q_1)
\right) e^{-i \omega_c(q_1) t} - e^{-i (2 \pi/L) v_{\rm F} t} \right]^l \\
a_{lm+i}^{(m)} (t) & = \sum_{j=0}^l \frac{ (1/m)^j }{j!} \Big[ \left( 1+\gamma_c(q_m) \right)
e^{-i \omega_c(q_m) t} \notag \\ & - e^{-i (2 \pi/L) m v_{\rm F} t} \Big]^j \ a_{m(l-j)+i}^{(m-1)} (t)
\end{align}
and
\begin{align}
& b_m^{(1)} (t) = \frac{1}{m!} \left[ \gamma_c(q_1) e^{-i \omega_c(q_1) t} \right]^m \\
& b_{lm+i}^{(m)} (t) = \sum_{j=0}^l \frac{ (1/m)^j }{j!} \left[ \gamma_c(q_m) e^{-i \omega_c(q_m) t} \right]^j
\ b_{m(l-j)+i}^{(m-1)} (t).
\end{align}
In terms of these we can rewrite
\begin{align}
i G^>_+(k_n, t) & = \frac{1}{2 \pi} \exp \left\{ -2 \sum_{n=1}^{\tilde{n}_{\rm c}} \frac{\gamma_c(q_n)}{n} \right\}
\notag \\ & \times
\sum_{r=\max \{0, -\tilde{n} \}}^{\infty} a_{ \tilde{n}+r }^{ (\tilde{n}_{\rm c}) } (t) \,
b_r^{ (\tilde{n}_{\rm c}) } (t)
\label{eq:g2g4spinless_G_coeff}
\end{align}
and the remaining Fourier transform to obtain $\rho^>_{+} (k_n, \omega)$ can be performed, again
after multiplication with $\exp\{-\chi |t| \}$, numerically by means of a FFT.
\subsection{Spectra of the spinful $g_4$-model}
\label{sect_g4}
Based on the formulas given in Sects.~\ref{sping4box} and \ref{sping4arb} we are in a position to compute
the exact spectral function of the spinful $g_4$-model for different potentials at finite system size $L$.
For the box potential both Fourier transforms can be performed analytically while for arbitrary
potentials the time-transform is performed numerically as a FFT. In this case only broadened spectra
can be obtained. We will compare the results to those derived from one of the ad hoc procedures
of Sect.~\ref{sect_adhoc}.
Besides the box potential we consider ($\nu=c,s$)
\begin{align}
g^{\rm p=4}_{4,\nu}(q) & = g_{4,\nu} \exp\{ -(q/q_{\rm c})^4/9\} \ & \text{ (p=4),} \label{p4pot} \\
g^{\rm gauss}_{4,\nu}(q) & = g_{4,\nu} \exp\{ -(q/q_{\rm c})^2 \} \ & \text{ (gauss),}
\label{eq:pot_gaus} \\
g^{\rm exp}_{4,\nu}(q) & = g_{4,\nu} \exp\{ -3 \, |q/q_{\rm c}| \} \ & \text{ (exp).}
\label{eq:pot_exp}
\end{align}
The factors in the exponential function were chosen such that, besides at $q=0$, all potentials have
the same value at $\tilde{q}_c = 3 q_c$; at this momentum they have decayed to $\approx 10^{-4}$
of the $q=0$ value and we can safely set the potentials to $0$ for $q > \tilde q_{\rm c}$.
Considering larger $\tilde{q}_c$ we have verified that this does indeed not affect our results.
For small
momenta the potentials go as $1-g_{4,\nu}(q)/ g_{4,\nu} \sim |q/q_{\rm c}|^p $, with $p=\infty$ for the box,
$p=4$ for the `p=4'-potential, $p=2$ for the Gaussian potential, and $p=1$ for the exponential potential.
The exponent $p$ is a measure for the flatness of the potential at $q=0$; see also Ref.~\onlinecite{Meden99}.
For definiteness in our calculations we have always chosen $g_{4,c}/(\pi v_{\rm F}) = 1/2 =-g_{4,s}
/ ( \pi v_{\rm F})$ without affecting our conclusions. In this case $v_c > v_s$. The measure
for the system size is $n_{\rm c}$. To numerically compute the recursively defined coefficients
within reasonable time we choose $n_{\rm c} = 5 \cdot 10^4$. The broadening $\chi$ is chosen such that
in the broadened spectral function no effects of the individual $\delta$-peaks are
visible. For the given $n_{\rm c}$
we take $\chi / (v_{\rm F} q_{\rm c}) = 5 \cdot 10^{-5}$.
\begin{figure}[h!]
\includegraphics[width=1.\linewidth,clip]{rho_g4spin_k001.png}
\includegraphics[width=1.\linewidth,clip]{rho_g4spin_k01.png}
\includegraphics[width=1.\linewidth,clip]{rho_g4spin_k03.png}
\caption{(Color online) Spectral function of the spinful $g_4$-model as a function
of energy for a) $(k-k_{\rm F})/q_{\rm c}=0.01$, b) $(k-k_{\rm F})/q_{\rm c}=0.1$, and c)
$(k-k_{\rm F})/q_{\rm c}=0.3$. Spectra for the different potentials are shown. In addition to
the FFT data with broadening $\chi / (v_{\rm F} q_{\rm c}) = 5 \cdot 10^{-5}$
we show unbroadened results for the box potential [`box (ana)'] and the broadened
results of the ad hoc procedure (iii) [`broadened (iii)']. The other parameters
are $g_{4,c}/(\pi v_{\rm F}) = 1/2 =-g_{4,s}
/ ( \pi v_{\rm F})$ and $n_{\rm c} = 5 \cdot 10^4$.
The curves are partly hidden by others (see the text).
The inset of c) shows the
position of the spin peak extracted from the data (stars) in comparison
to the collective spin mode dispersion $\omega_s(k-k_{\rm F})$ (lines). The dashed
line displays the unrenormalized dispersion $v_{\rm F}(k-k_{\rm F})$.
}
\label{fig1}
\end{figure}
In Figs.~\ref{fig1}a) to \ref{fig1}c) we show $\rho^>_{+} (k, \omega)$ for three different
small $(k-k_{\rm F})/q_{\rm c}$. The curves labeled as `box (ana)' are the results for the box
potential obtained from Eq.~\eqref{eq:g4spinrhoex} without convoluting it with a
Lorentzian. The weights of the $\delta$-peaks are divided by the level spacing (which for
the given parameters is $1/n_{\rm c}$) and are connected to form a smooth curve. All other results
(up to `broadened (iii)'; see below) were obtained from Eq.~\eqref{eq:g4spinG(k,t)}
multiplied with an exponentially decaying function $\exp\{-\chi |t| \}$ and transformed with
a FFT; for comparison we also show the broadened spectra for the box potential [box (FFT)].
As can be seen
from Fig.~\ref{fig1}a), for $k$ very close to $k_{\rm F}$
all curves are nearly indistinguishable. The broadening of the spectra is visible by the weight
`leaking out' for $\omega < v_s [k-k_{\rm F}] $ and $\omega > v_c [k-k_{\rm F}]$. However,
already for $(k-k_{\rm F})/q_{\rm c} = 0.1$ [Fig.~\ref{fig1}b)], the spectral function calculated
with an exponentially decaying potential shows pronounced differences to the other ones.
For $(k-k_F)/q_c = 0.3$, also the curve of the Gaussian potential deviates from the one of the box
potential, see Fig.~\ref{fig1}c). The spectrum of the `very flat' p=4 potential still lies
on top of the one obtained for the box potential; differences only appear at even larger
$k-k_{\rm F}$ (not shown).
Obviously, the smaller the above introduced index $p$, that is the less `flat' the potential is
at $q=0$, the faster the line shape of the spectra starts to deviate from the one of the box potential
when $k- k_{\rm F}$ increases.
The maxima of $\rho^>_+(k,\omega)$ are located at
$\omega_{\nu}(k-k_{\rm F})$, instead of at $v_\nu [k-k_{\rm F}] $; for $|k-k_{\rm F}|< q_{\rm c}$
both positions are equal for the box potential.
This is shown in the inset of Fig.~\ref{fig1}c), where the spin dispersion relation for the
different potentials (full lines) is compared to the numerically determined maxima of the spectra
(stars). The agreement is very good; the charge peak behaves similarly. In the limit
of small $k-k_{\rm F}$ and large $p$ the difference between $\omega_{\nu}(k-k_{\rm F})$ and
$ v_\nu [k-k_{\rm F}]$ is negligible.
In the literature the third ad hoc regularization procedure leading to
$[G_{+}^{>}]_{\rm (iii)} (x,t)$ Eq.~(\ref{gapprox3}) is considered `the best' one\cite{Voit93b,Voit95}
as it leads to a $[\rho^>_+]_{\rm (iii)}(k,\omega)$ which fulfills exact sum rules.\cite{Suzumura80,Schoenhammer93b}
For $\gamma_c=\gamma_s=0$ a closed analytical expression of the double Fourier transform
can be given; see Eqs.~(3.20) and (3.21) of Ref.~\onlinecite{Voit93b}.
We therefore added a graph of this analytical result convoluted with a Lorentzian of width $\chi$
as the dashed lines in Figs.~\ref{fig1}a) to \ref{fig1}c) (with $1/\Lambda \to q_{\rm c}$).
As mentioned above these curves as well as the exact function for the box potential show single-sided
(threshold) square-root singularities at $v_{\nu} [k-k_{\rm F}]$ (for a more detailed analysis on this see below).
The approximate spectrum $[\rho^>_+]_{\rm (iii)}(k,\omega)$ is characterized by an
additional feature at $\omega= v_{\rm F} [k-k_{\rm F}]$ (barely visible in Fig.~\ref{fig1}a)
for $(k-k_{\rm F})/q_{\rm c}=0.01$); as analyzed in Ref.~\onlinecite{Voit93b} a logarithmic
divergence appears at this energy. It results from the additional factor
$(x- v_{\rm F} t + i \Lambda)/(x- v_{\rm F} t +i0^+)$ of $[G_{+}^{>}]_{\rm (iii)} (x,t)$
Eq.~(\ref{gapprox3}). Obviously, the exact spectral function of the $g_4$-model does not
display this feature for any of the considered potentials. For the box potential, which besides
the missing peak at $\omega= v_{\rm F} [k-k_{\rm F}]$ shows the same spectral characteristics
as found in the ad hoc procedure, this can even be understood analytically: as seen
in Eq.~(\ref{eq:g4spinrhoex}) for $k-k_{\rm F} < q_{\rm c}$ the information on $v_{\rm F}$
drops completely out. The logarithmic divergence
of $[\rho^>_+]_{\rm (iii)}(k,\omega)$ at $\omega= v_{\rm F} [k-k_{\rm F}]$ is thus an {\it artifact
of the ad hoc regularization.} As discussed very recently, this logarithmic divergence
for $k-k_F \lessapprox q_{\rm c}$ turns into a
power-law one if the spinful ($g_2$-$g_4$) TLM is treated within the ad hoc procedure
(iii).\cite{Maebashi14} Also this feature is an artifact of the ad hoc procedure.
With Ref.~\onlinecite{Maebashi14} in mind we emphasize that this does not exclude
that for $k-k_{\rm F} \gg q_{\rm c}$ all the spectral weight is located around $\omega =
v_{\rm F} (k-k_{\rm F})$, which, in fact, is generically
the case. This was discussed for the box potential in Ref.~\onlinecite{Schoenhammer93} and for
general interactions in Ref.~\onlinecite{Schoenhammer93b}. We here do not investigate this any
further as we are exclusively interested in the spectra at small $|k-k_{\rm F}|$.
In order to investigate whether or not the spectra at fixed $0< k-k_{\rm F} \ll q_{\rm c}$ and
for $\omega$ close to the maxima at $\omega_{\rm max}$
show power-law scaling we do not
simply want to rely on the quality of power-law fits to the broadened data. Instead we take
the logarithmic derivative
\begin{equation}
\mbox{logder}(\omega)=\frac{ d \ln \left[ \rho^>_{+} (k, \omega) \right] }{d \ln | \omega- \omega_{\rm max} | },
\label{eq:logder}
\end{equation}
of our data with $\omega_{\rm max}$ equal to the corresponding peak positions. It is numerically approximated by
centered differences and provides a very sensitive approach in the search for power laws.
If for $\omega \to \omega_{\rm max}$, $\mbox{logder}(\omega)$ approaches a constant $\xi$
the spectral function shows power-law scaling with exponent $\xi$ close to $\omega_{\rm max}$ (according
to the definition given in the introduction).
\begin{figure}[th!]
\includegraphics[width=1\linewidth,clip]{logder_g4spin_k001.png}
\includegraphics[width=1\linewidth,clip]{logder_g4spin_k01.png}
\caption{(Color online) Logarithmic derivative Eq.~(\ref{eq:logder}) of the
spectral function of the spinful $g_4$-model close to the spin peak
for a) $(k-k_{\rm F})/q_{\rm c}=0.01$ and b) $(k-k_{\rm F})/q_{\rm c}=0.1$. In addition to
the broadened FFT data for the different potentials and the
unbroadened results for the box potential [`box (ana)']
we show broadened [`broadened pl'] as well as unbroadened
[`pl'] data for the product of single-sided square-root singularities Eq.~(\ref{eq:g4powlaw}).
The parameters are as in Fig.~\ref{fig1}.
}
\label{fig2}
\end{figure}
In Figs.~\ref{fig2}a) [$(k-k_{\rm F})/q_{\rm c}=0.01$] and \ref{fig2}b) [$(k-k_{\rm F})/q_{\rm c}=0.1$]
we show $\mbox{logder}(\omega)$ close to the spin peak for the different
two-particle potentials. For symmetry reasons the behavior close to the charge peak is the same.
Instead of an ad hoc regularized spectral function -- which is spoiled by the spurious peak at
$\omega = v_{\rm F} (k-k_{\rm F})$ -- we this time present the logarithmic derivative
of the simple normalized product of two single-sided square-root singularities
\begin{equation}
[\rho^>_+]_{\mathrm{pl}} (k,\omega) = \frac{1}{\pi} \, \frac{ \Theta
\left( \omega - v_s [k-k_F] \right) }{ \left( \omega - v_s [k-k_F] \right)^{1/2} } \,
\frac{ \Theta \left( v_c [k-k_F] - \omega \right) }{ \left( v_c [k-k_F] - \omega \right)^{1/2} },
\label{eq:g4powlaw}
\end{equation}
for reference. It is indicated by `pl'. In addition we present the logarithmic derivative of this
expression convoluted with a Lorentzian of width $\chi$, indicated by `broadened pl'.
The unbroadened data for the box potential -- for which we analytically know that for $L \to \infty$
a square root singularity at $v_s [k-k_{\rm F}]$ exists [see Eq.~(\ref{wurz})] -- very nicely follow the `pl' curve
down to $\omega - \omega_{\rm max} \approx 4 \cdot 10^{-4} v_{\rm F} q_{\rm c}$.
At this energy {\it finite size corrections}
destroy the power-law scaling. This can be verified by considering different $n_{\rm c}$, that is
different system sizes. The `broadened pl' curve starts to deviate from the unbroadened one
at $\omega - \omega_{\rm max} \approx 6 \cdot 10^{-4} v_{\rm F} q_{\rm c} $.
From this we conclude that for the chosen $n_{\rm c}$
and $\chi$ the broadening $\chi$ cuts off the power-law scaling at slightly larger energies
than the system size.
This is consistent with the observation that the broadened spectrum of the box potential [`box (FFT)']
almost perfectly follows the `broadened pl' curve down to the much smaller scale
$\omega - \omega_{\rm max} \approx 6 \cdot 10^{-5} v_{\rm F} q_{\rm c}$. This gives us confidence that the FFT data for
the other potentials are unspoiled by both finite size and broadening effects down to
$\omega - \omega_{\rm max} \approx 6 \cdot 10^{-4} v_{\rm F} q_{\rm c}$. At the `high energy' end possible power law
scaling close to the spin peak is cutoff by the charge peak. The latter is the origin of the
dominant feature at $\omega-\omega_{\rm max} \approx 10^{-2} v_{\rm F} q_{\rm c}$ and
$\omega-\omega_{\rm max} \approx 10^{-1} v_{\rm F} q_{\rm c}$, respectively.
While for $(k-k_{\rm F})/q_{\rm c}=0.01$ the data of the `p=4' and the `gauss' potential
lie perfectly on top of the `box (FFT)' data, and one is tempted to conclude that they are consistent with a
square-root singularity at $\omega_{\rm max}$, the data for the exponential potential clearly differ
and are inconsistent with such behavior. For $(k-k_{\rm F})/q_{\rm c}=0.1$ in addition
the data for the Gaussian potential are incompatible with this type of singularity. For even
larger $k-k_{\rm F}$ (but still smaller than $q_{\rm c}$; not shown) also the data for the
`p=4' potential no longer follow the ones of the box potential.
The most consistent interpretation of our results is that for any potential which is not
`infinitely flat' at $q=0$, that is
if $p<\infty$, the spectral function at fixed $k-k_{\rm F} >0$ does strictly speaking not show
power-law scaling close to $\omega_{\rm max}$. The less `flat' the potential is, that
is, the smaller $p$, the faster this becomes apparent as $k-k_{\rm F}$ is increased. The
square-root singularities found for the box potential (and for the ad hoc procedures) are cut
off by the curvature of the potential close to $q=0$.
They can thus not be considered as universal features of the spinful $g_4$-model.
We note that the data for $p<\infty$ are not only inconsistent with power-law scaling
when taking $\omega_{\rm max}$ as the point of reference. We studied the behavior relative to
other distinguished energies
(e.g. $v_\nu [k-k_{\rm F}]$ and the -- due to the broadening -- apparent thresholds). For
none of these we find behavior which is consistent with power laws.
Despite this lack of power-law behavior, for all interaction potentials studied
the exact spectral function of the $g_4$-model is still characterized by spin and
charge peaks.
\subsection{Spectra of the spinless $g_2$-$g_4$-model}
\label{sect_g2_g4}
Employing the formulas given in Sects.~\ref{nospinbox} and \ref{nospinarb} we can compute
the exact spectral function of the spinless TLM -- the spinless $g_2$-$g_4$-model -- for
different forms of the potential. In the last section we saw that for the small
$k-k_{\rm F}$ we are interested in, the spectra for the `p=4' potential Eq.~(\ref{p4pot})
barely differ from the ones obtained for the box potential. The same holds for the full
spinless model and in this section we focus on the box potential, the Gaussian one Eq.~(\ref{eq:pot_gaus})
as well as the exponential potential Eq.~(\ref{eq:pot_exp}). To prevent an inflation of
cases and parameters we consider the physically reasonable situation of equal inter- and
intra-branch scattering $g_2(q) = g_4(q) = g(q)$. We choose $g(0)$ such that $v_c=2 v_{\rm F}$ and $\gamma_c(0)=1/8$.
To compute $\rho^>_+(k,\omega)$ via $G^>_+(k,t)$ Eq.~(\ref{eq:g2g4spinless_G_coeff})
and FFT for general potentials or via Eq.~(\ref{eq:g2g4rho_boxex}) for the box potential
we have to recursively compute more coefficients and perform additional sums in comparison
to what was necessary in the spinless $g_4$-model. This increases
the numerical resources required and we thus have to consider smaller system sizes compared
to the latter; we choose $n_{\rm c} = 2 \cdot 10^4$. Furthermore, at fixed $n_{\rm c}$
the energy level spacing of the spinless $g_2$-$g_4$-model is larger than that of the spinful
$g_4$-model. To obtain smooth curves we thus have to increase the broadening $\chi$. We take
$\chi/(v_{\rm F} q_{\rm c}) = 10^{-3}$.
In Figs.~\ref{fig3}a) to \ref{fig3}c) we present the {\it total spectral function}
$\rho_+(k,\omega) = \rho^>_{+} (k, \omega) + \rho^<_{+} (k, \omega)$ for $(k-k_{\rm F})/q_{\rm c}=0$, $0.01$,
and $0.1$. We switched to this as it simultaneously shows the photoemission as well
as inverse photoemission part of the spectrum. For increasing $k-k_{\rm F }$ the photoemission
part looses weight quickly. Thus in Fig.~\ref{fig3}c) with $(k-k_{\rm F})/q_{\rm c}=0.1$ we only
present a zoom-in of the inverse photoemission part. In addition to the broadened functions for
the box, Gaussian, and exponential potentials obtained by FFT we show the unbroadened one of the
box potential [see Eq.~(\ref{eq:g2g4rho_boxex})]. As above, to obtain the latter the weights of
the $\delta$-peaks were divided by the level spacing (which for the given parameters is $4/n_{\rm c}$)
and are connected to form a smooth curve.
\begin{figure}[h!]
\includegraphics[width=1.07\linewidth,clip]{rho_g2g4spinless_k0.png}
\includegraphics[width=1.07\linewidth,clip]{rho_g2g4spinless_k001.png}
\includegraphics[width=1.07\linewidth,clip]{rho_g2g4spinless_k01.png}
\caption{(Color online) Total spectral function of the $g_2$-$g_4$-model as a function
of energy for a) $(k-k_{\rm F})/q_{\rm c}=0$, b) $(k-k_{\rm F})/q_{\rm c}=0.01$, and c)
$(k-k_{\rm F})/q_{\rm c}=0.1$. Spectra for the different potentials are shown. In addition to
the FFT data with broadening $\chi / (v_{\rm F} q_{\rm c}) = 10^{-3}$
we show unbroadened results for the box potential [`box (ana)']. The
interaction at $q=0$ is chosen such that $v_c = 2 v_{\rm F}$ and $\gamma_c(0)=1/8$. The
system size is set by $n_{\rm c} = 2 \cdot 10^4$.
}
\label{fig3}
\end{figure}
As discussed in Sect.~\ref{nospinbox} for $0< k-k_{\rm F} < q_{\rm c}$ and the box potential
$\rho_+(k,\omega)$ shows threshold power-law nonanalyticities at $\pm v_c [k-k_{\rm F}]$ with
exponents $\gamma_c-1$ (for $\omega>0$) and $\gamma_c$ (for $\omega <0$). In `box (FFT)' these
are broadened. Similarly to the spinful $g_4$-model we observe that the smaller $p$ the faster
the line shape starts to deviate from the one of the box potential when increasing $k-k_{\rm F}$.
For $(k-k_{\rm F})/q_{\rm c}=0.1$ [see Fig.~\ref{fig3}c)] and the exponential potential with
$p=1$ this already leads to a strongly modified distribution of the spectral weight. The
deformed line shape can be understood in due detail when comparing it to the spectral function
of the {\it spinless} $g_4$-model.\cite{Markhof15} This detailed analysis is beyond the
scope of the present paper.
From Ref.~\onlinecite{Meden99} and the RG irrelevance
of the momentum dependence of the two-particle potential we expect that for all potentials
$\rho_+(k_{\rm F},\omega) \sim |\omega|^{2 \gamma_c -1}$. On first glance the data of Fig.~\ref{fig3}a)
appear to be consistent with this behavior, however, a more thorough analysis is required.
In Fig.~\ref{fig4}a) we plot the logarithmic derivative Eq.~(\ref{eq:logder}) (with $\omega_{\rm max}=0$)
of the $\omega>0$ broadened FFT data of Fig.~\ref{fig3}a) as dotted lines. While the logarithmic
derivative of the unbroadened `box (ana)' data (solid line) nicely shows a plateau at the
expected exponent $2 \gamma_c -1 = -3/4$ (for the given parameters), which is only spoiled at
very small $\omega/(v_{\rm F} q_{\rm c}) \approx 10^{-3}$ due to finite size effects, the dotted
curves do not seem to support power-law scaling of $\rho_+(k_{\rm F},\omega)$. This also holds
for the {\it broadened} `box (FFT)' data which establishes that the {\it broadening destroys the
power law even for} $\omega \gg \chi$.
To further analyze this surprising finding in a `controlled' setup we took the function $f(x) = x^{-3/4}$
and convoluted it with a Lorentzian of width $10^{-3}$. We indeed found that for $x \in [10^{-2},10^{-1}]$
the logarithmic derivative of the resulting function shows a behavior quite similar to the one of the
dotted lines in Fig.~\ref{fig4}a); in particular, it bends away from the expected plateau towards smaller
values. For larger exponents, let's say -1/2 as in Figs.~\ref{fig2}a) and b), this does not happen. One is
thus tempted to increase the interaction strength and thus $2 \gamma_c -1$. However, larger interactions
also imply larger energy level spacings of the $\delta$-peaks and thus require larger $\chi$, which cuts off
the power-law scaling at larger energies.
As a possible way out we performed a {\it deconvolution} of our numerically obtained spectral function along
the lines of Refs.~\onlinecite{Gebhard03} and \onlinecite{Raas05}. The resolution was chosen to be
approximately equal to the broadening $\chi$. The deconvolution of numerical data is of course an
ill-posed problem and the corresponding spectral function can e.g. become negative.\cite{Raas05}
We were able to perform a stable deconvolution down to $(\omega-\omega_{\rm max})/(v_{\rm F} q_{\rm c})
\approx 10^{-2}$. In fact, the resulting spectra are {\it sufficiently smooth to perform
logarithmic derivatives.} These are shown as solid lines in Fig.~\ref{fig4}a). The oscillatory
behavior at the lower end signals the onset of an instability of the deconvolution. The deconvoluted
data for the box potential lie exactly on top of the ones obtained from the unbroadened spectral
function `box (ana)'. This gives us confidence that also for the other potentials the deconvolution
can be trusted. This is further supported by the observation that the data for the Gaussian potential
now clearly support power-law scaling with exponent $2 \gamma_c -1$. For the exponential potential the energy
scale cutting off the power-law scaling at the {\it `high energy' side} appears to be of the order
$10^{-2}$ and thus smaller compared to the one of the box (of the order $1$) and the Gaussian potential
(of the order $10^{-1}$). Therefore no clear plateau is reached for the accessible energies.
However, the data appear to saturate at the expected value $2 \gamma_{c}-1$.
\begin{figure}[h!]
\includegraphics[width=1.\linewidth,clip]{logder_g2g4spinless_k0.png}
\includegraphics[width=1.\linewidth,clip]{logder_g2g4spinless_k001.png}
\includegraphics[width=1.\linewidth,clip]{logder_g2g4spinless_k01.png}
\caption{(Color online)
Logarithmic derivative Eq.~(\ref{eq:logder}) of the
spectral function of the spinless $g_2$-$g_4$-model close to the
inverse photoemission peak
for a) $(k-k_{\rm F})/q_{\rm c}=0$ and b) $(k-k_{\rm F})/q_{\rm c}=0.01$, and
c) $(k-k_{\rm F})/q_{\rm c}=0.1$.
The full lines labeled by `box (ana)' show the unbroadened results of the box
potential. The dotted lines in a) result from broadened spectra. The other
full lines are obtained from the numerical spectra after a deconvolution
(see the text). The dashed lines indicate the expected exponent.
The parameters are as in Fig.~\ref{fig3}.
}
\label{fig4}
\end{figure}
Along the same lines we next investigate whether or not for $k-k_{\rm F} > 0$ a power law is found
close to the peak on the inverse photoemission side. In Figs.~\ref{fig4}b)
[$(k-k_{\rm F})/q_{\rm c}=0.01$] and c) [$(k-k_{\rm F})/q_{\rm c}=0.1$] we
present the logarithmic derivative of the deconvoluted spectra together with `box (ana)' data.
The logarithmic derivative of the original spectra behave similar to the $k=k_{\rm F}$ case and are
thus not shown. As expected the `box (ana)' data are consistent with power-law scaling with exponent
$\gamma_c-1 = -7/8$ (for the given parameters). The power law is cut off at $(\omega-\omega_{\rm max})
/(v_{\rm F} q_{\rm c}) \approx 10^{-3}$ due to finite size effects. The deconvoluted box potential data
fall again exactly on top of the `box (ana)' ones indicating that the deconvolution is stable.
For $(k-k_{\rm F})/q_{\rm c}=0.01$ the data for the Gaussian potential are for sufficiently small
$\omega-\omega_{\rm max}$ on top of the box potential ones [see Fig.~\ref{fig4}b)].
This does no longer hold for $(k-k_{\rm F})/q_{\rm c}=0.1$ [see Fig.~\ref{fig4}c)]
for which the data of the Gaussian potential are incompatible with power-law scaling.
For the exponential potential with smaller $p$ this deviation from possible power-law behavior
sets in at already smaller $k-k_{\rm F}$ [see Fig.~\ref{fig4}b)].
As for the spinful $g_4$-model
the most consistent interpretation of our results is that for any potential with $p<\infty$
and fixed $k-k_{\rm F} \neq 0$ the power-law scaling found for the box potential is destroyed by
the curvature of the potential at $q=0$; power laws are thus nongeneric.
We note that this does not only hold when taking $\omega_{\rm max}$ as the
point of reference. We studied the behavior relative to other distinguished energies
(e.g. $v_c [k-k_{\rm F}]$ and the -- due to the broadening -- apparent threshold). For none of these
points we find behavior which is consistent with power laws. Similarly also the behavior close
to the threshold on the photoemission side are incompatible with power-law scaling if $p<\infty$.
These results imply that the spectral functions resulting from the different ad hoc
regularizations (not shown here), which are all characterized by the threshold power laws
Eqs.~(\ref{power1}) and (\ref{power2}), are nongeneric.
\section{Discussion}
\label{sect_discussion}
Our above results for the exact spectral function can be summarized as follows:
\begin{enumerate}
\item For $k-k_{\rm F} \neq 0$ and generic two-particle potentials which are not `infinitely flat'
at momentum transfer $q=0$ the spectral function does not show power-law scaling close to any
of the distinguished energies. Power-law behavior is generically only found if all energy scales are sent
to zero, e.g. in $\rho^>(k_{\rm F},\omega)$ for $\omega \to 0$ (and if $g_2(0)$ is finite).
\item The ad hoc regularized spectra which are commonly studied show finite $k-k_{\rm F}$ power laws
and can thus not be considered as generic. The origin of this nongeneric behavior is the linearization
of the spin and charge dispersion relations.
The ad hoc spectra are plagued by additional artifacts (see Sect.~\ref{sect_g4}).
\item The less flat the potential at $q=0$, that is the smaller the introduced index $p$, the
faster the differences of the spectral line shape compared to the one of the box potential --
and in many respects compared to the one of the ad hoc procedures --
becomes apparent when $k-k_{\rm F}$ is increased. For small $p$, e.g. the exponential potential
with $p=1$, already for $(k-k_{\rm F})/q_{\rm c}=0.1$ major differences are apparent; see Figs.~\ref{fig1}b)
and \ref{fig3}c). For $k-k_{\rm F} \neq 0$ the spectral function is still characterized by the
dispersing spin and charge peaks.
\end{enumerate}
To individually study how the curvature of the potential at zero momentum transfer modifies
the two interaction effects of spin-charge separation and power laws with
interaction dependent exponents, we have studied the spinful $g_4$-model and the spinless
$g_2$-$g_4$-model. Combining the two limiting cases it is obvious that the same conclusions
can be drawn for the full TLM, that is the spinful $g_2$-$g_4$-model. Explicit results
for the spectral function of the full TLM are presented in Sect.~\ref{sect_con_exp} (see
Fig.~\ref{fig5}).
We next discuss the implications of these findings.
\subsection{Luttinger liquid universality}
\label{sect_con_LLun}
\subsubsection{General considerations}
The TLM forms the low-energy fixed point model under RG flow of a large class of 1d correlated
fermion models.\cite{Solyom79} Based on this insight it was suggested that the power laws of
the ad hoc regularized TLM spectral function $\rho_+^>(k \neq k_{\rm F},\omega)$ should be
observable in other models from this class.\cite{Voit95} One can doubt this on general grounds
as $k-k_{\rm F} \neq 0$ sets a finite scale thus breaking quantum critical scale invariance
and cutting off the RG flow.\cite{Meden99} Our results show {\it explicitly} that power laws in
$\rho^>(k \neq k_{\rm F},\omega)$ are indeed not part of LL universality; if they cannot be
found in the generic TLM, they cannot expected to be a universal feature of other models.
We emphasize that LL universality does {\it not} imply that for a given model from the LL
universality class (other than the TLM itself) one simply has to choose proper
coupling functions $g_{i,\kappa}(q)$ -- which in any case would be unknown a priori --
and can reproduce details of the spectral function of this model at low energies using
the TLM. All that is known about the spectral properties of a model from the LL universality
class is that for $\omega \to 0^+$ the scaling relations $\rho^>(k_{\rm F},\omega) \sim
\omega^{\alpha-1}$ and $\rho^>(\omega) \sim \omega^\alpha$ hold.
When computing space-time correlation functions of the TLM other than the single-particle
Green function momentum integrals of similar type as encountered here appear. To evaluate
these ad hoc procedures including the linearization of the collective spin and
charge dispersion are commonly employed.\cite{Voit95} One prominent example is the
density-density correlation function for momentum transfer close to $2 k_{\rm F}$.\cite{Luther74}
As for the single-particle spectral function after regularization the Fourier integrals
can be performed analytically leading to power-law scaling of the susceptibility with
interaction dependent exponents close to the characteristic energies
$\pm v_c [k \pm 2 k_{\rm F}]$. Our considerations imply that also this feature can most
likely not be considered as being characteristic for LLs in general. The density response
at {\it vanishing energy} and for $k \to \pm 2 k_{\rm F}$ (all energy scales are sent to 0)
is, however, in general characterized by a power-law divergence (for repulsive interactions)
which indicates the breakdown of linear
response theory. It is a signature of the sensitivity of a LL against single-particle
perturbations with momentum transfer $2 k_{\rm F}$ (backscattering).\cite{Giamarchi03,Schoenhammer05}
We note that the character of the density-density correlation function with small momentum
transfer is not modified by the momentum dependence of the potentials; it is
given by a $\delta$-function at energy $\omega_c(q)$ instead of $v_c |q|$.
\subsubsection{Spectra of lattice models}
Directly computing the single-particle spectral function for translational
invariant microscopic lattice models of 1d correlated fermions still poses a
formidable challenge of quantum many-body theory. Two promising routes exists.
The first one is numerical in nature. Using exact diagonalization (ED),
the (dynamical) density-matrix RG (DMRG), or different types of quantum Monte-Carlo (QMC) approaches,
valuable information on the spectral function of different models was collected. However,
the search for power laws requires an exceptional energy resolution. This implies that fairly
large system sizes and low temperatures must be accessible. The model most heavily investigated
is the 1d Hubbard model (and variants of it). Away from half-filling it is known
to fall into the LL universality class. ED is restricted to systems of a few ten lattice
sites which leads to a poor energy resolution. Within so-called
cluster perturbation theory, which is ED-based, it was at least possible to observe spin-charge
separation.\cite{Senechal00} The latter was systematically studied using
QMC.\cite{Zacher98,Abendschein06} In these studies the finite temperature (and the required analytic continuation)
turned out to be the main obstacle preventing an analysis of
$\rho^>(k ,\omega)$ in terms of power-law scaling. Reference \onlinecite{Benthien04}
contains the first serious attempt in this direction. The authors use (dynamical) DMRG to obtain
broadened spectra. A scaling analysis as a function of the broadening was interpreted to be consistent
with power-law behavior of the charge and spin peaks. The quality of the data is, however, not
good enough to either confirm or refute power laws at $k-k_{\rm F} \neq 0$.
Generally speaking the numerical results for lattice models from the LL universality class
are fully consistent with our conclusions.
A word of warning is in order. As our analysis of Sect.~\ref{sect_g2_g4} shows
it can be very difficult to undoubtedly confirm or refute power-law behavior of broadened
finite size data. This holds even though for the TLM we can achieve a comparably high
energy resolution and obtain data which are sufficiently accurate to employ a very sensitive
logarithmic derivative.
A promising analytical approach to the spectral function of 1d models is built upon the special
structure of several of the standard lattice models from the LL universality class, namely the
existence of an extensive number of local integrals of motion. For this reason several models
are exactly solvable by Bethe ansatz, which, however, does not imply that the single-particle
spectral function can (easily) be computed exactly.\cite{Essler05} In a series of papers,
see Ref.~\onlinecite{Carmelo06} and references therein, $\rho^>(k ,\omega)$ of the 1d Hubbard
model was investigated using a `pseudo\-fermion dynamical theory' which is based on the Bethe
ansatz solution. It was reported that the spectral function contains power laws even for
$k - k_{\rm F} \neq 0$, with exponents which depend on $k - k_{\rm F}$. This finding
might be related to the `nonlinear Luttinger liquid phenomenology' briefly touched below.
The spectral function of another exactly solvable 1d model, namely the Calogero-Sutherland
model showing power laws with momentum dependent exponents was interpreted in the light of
this phenomenology.\cite{Khodas07}
%The computation of $\rho^>(k ,\omega)$ for this model
%is built upon a continuation from integer coupling constants to arbitrary ones which lacks
%a detailed justification.
We emphasize that our findings do not exclude that specific models from the LL
universality class might show finite $k - k_{\rm F}$ power laws, however, if so for
{\it more specific reasons than LL universality.}
Models with equilibrium dynamics which are restricted by an extensive
number of local conserved quantities might be examples for such behavior.
This is also supported by the observation that
the power laws reported on in Refs.~\onlinecite{Benthien04,Carmelo06,Khodas07}
cannot only be found at low energies (for very small $|k-k_{\rm F}|$), while
LL theory is supposed to be applicable in this limit only.
\subsection{Implications for the `nonlinear Luttinger liquid phenomenology'}
\label{sect_con_nonlinLL}
In an extensive series of papers, reviewed in Ref.~\onlinecite{Imambekov12},
a novel approach to study corrections to LL behavior by the curvature
of the single-particle dispersion was developed. It cumulates in the so-called
`nonlinear Luttinger liquid phenomenology'. The approach is mainly built upon
an effective field theory which is motivated by lowest order perturbation
theory. For the single-particle spectral function this phenomenology predicts
(Fermi edge singularity like) power laws with momentum dependent exponents even at
$k-k_{\rm F} \neq 0$. However, the field theory relies on the assumption
of a momentum independent bulk interaction (the momentum dependence of the
interaction with the mobile impurity is considered) and requires ad hoc
ultraviolet regularization. As we have shown in (linear) Luttinger liquid
theory the same assumption leads to power-law behavior which is nongeneric
rather than universal. This raises the question whether the power laws found in
`nonlinear LL phenomenology' are robust against a curvature of the bulk
two-particle potential.
\subsection{Implications for the interpretation of experimental spectra}
\label{sect_con_exp}
\begin{figure}[t]
\includegraphics[width=1.\linewidth,clip]{rho_g2g4spin_gaus.png}
\includegraphics[width=1.\linewidth,clip]{rho_g2g4spin_exp.png}
\caption{(Color online) Photoemission part of the spectral function of
the spinful TLM at $(k-k_{\rm F})/q_{\rm c}=-0.1$. A spin-independent
interaction with equal inter- and intra-branch amplitude is
assumed. The interaction strength is chosen such that $\gamma_c(q=0)$ assumes
the values given in the legend. The other parameters are
$n_{\rm c} =10^3$ (system size) and $\chi/(v_{\rm F} q_{\rm c})=3 \cdot 10^{-3}$ (broadening).
In a) the spectra for the Gaussian
potential are shown, in b) the ones for the exponential
potential. The small `high-energy' oscillations for larger interactions
in a) are a finite size effect.
}
\label{fig5}
\end{figure}
When interpreting experimental angular resolved photoemission data on
quasi-1d metallic materials\cite{Grioni09} the ad hoc regularized spectral function of
the TLM is often taken paradigmatically. It is e.g. expected that the distribution
of the spectral weight between the spin and charge peaks as well as the spacing
between the two must be exactly as in the analytical expressions given in
Refs.~\onlinecite{Meden92,Voit93,Voit93b,Voit95}. A very recent example
in which this leads to a putative conflict can be found in Ref.~\onlinecite{Ohtsubo15}.
Our results for the spinful $g_4$- and the spinless $g_2$-$g_4$-models
show that the ad hoc regularized spectral function cannot be
considered as universal. Within the TLM the details of the distribution of spectral
weight, the peak distance, and the line shape clearly depend on the form of the
two-particle potential even when considering small $|k-k_{\rm F}|$. Such `details' are
not part of LL universality and {\it expecting quantitative agreement
overstresses LL theory.} We reiterate that all that is known about the $T=0$ spectral
properties from the latter is that for $\omega \to 0^+$ the scaling relations
$\rho^>(k_{\rm F},\omega) \sim \omega^{\alpha-1}$ and $\rho^>(\omega) \sim \omega^\alpha$ hold.
The exponent $\alpha=\gamma_c+\gamma_s$ can be expressed in terms of the LL parameters
$K_{c/s}$ [see Eq.~(\ref{gammadef}); $K_s =1$ for spin-rotational invariant systems],
which also enter in other `observables'.\cite{Giamarchi03,Schoenhammer05} If they can
be measured for the same system, consistency checks are possible. LL universality
also makes predictions for the scaling of spectral weight as a function $T$. It
was shown that $\rho^>(\omega=0) \sim T^\alpha$ and $\rho^>(k_{\rm F},\omega=0)
\sim T^{\alpha-1}$.\cite{Schoenhammer93c,Nakamura97,Orgad01} We note that details
of the analysis of $\rho_+^>(k,\omega)$ at $T>0$ for the TLM\cite{Nakamura97,Orgad01}
beyond the above scaling relation should be taken with caution as they rely
on an ad hoc regularization procedure.
It is tempting to employ the (spinful) TLM spectra computed with proper
$g_{i,\kappa}(q)$ for comparison to experimental ones beyond the above scaling
relations. This might lead to a {\it qualitative agreement} of certain features.
One crucial generic feature we found regardless of the $q$-dependence of the
two-particle potential are dispersing spin and charge `peaks', however,
generically not given by power-law singularities. To illustrate this
we show the spectral function of the spinful TLM for the Gaussian and the
exponential potential in Figs.~\ref{fig5}a) and b), respectively. This time we
focus on the {\it photoemission part} $\rho_+^<(k,\omega)$ for $(k-k_{\rm F})/q_{\rm c} = -0.1$
as this is most easily accessible experimentally. The results were obtained by a
straightforward generalization
of the recursive procedure put forward for the spinful $g_4$-model
(see Sect.~\ref{sping4arb}) and the spinless $g_2$-$g_4$-model
(see Sect.~\ref{nospinarb}). The system size is set by $n_{\rm c} = 10^3$ and
the broadening by $\chi/(v_{\rm F} q_{\rm c})=3 \cdot 10^{-3}$. We consider the physically reasonable case
of a spin-independent interaction with an equal amplitude of the intra- and inter-branch
parts. This implies $\omega_s(q) = v_{\rm F} q$ and thus $v_s(q=0) = v_{\rm F}$.
The interaction is varied such that $\gamma_c(q=0)=1/8$, $1/3$, and $9/16$.
Accordingly, $v_c(q=0)$ increases. As for the above discussed limiting cases
of the spinful $g_4$- and the spinless $g_2$-$g_4$-model, the details of the spectral
line shape obviously depend on the details of the two-particle potential
[compare Figs.~\ref{fig5}a) and b)]. The integral over $\rho_+(k,\omega) =
\rho^>_{+} (k, \omega) + \rho^<_{+} (k, \omega)$ at fixed $k-k_{\rm F}$ must be
normalized to 1 which implies that for the exponential potential less spectral
weight is transferred to $ \rho^>_{+} (k, \omega)$ (not shown) than for the Gaussian one.
For the former the `charge peak' is fairly broad and has an unusal line shape.
However, the two sets of spectra also share similarities. For increasing interaction
the spin peak loses weight and deforms into a shoulder like
structure. To summarize, we expect spin-charge separation to be a robust feature
of quasi-1d metalls which should be observable in photoemission experiments.
\begin{acknowledgements}
We are grateful to Kurt Sch\"onhammer for numerous discussions. This work was supported
by the DFG via RTG 1995.
\end{acknowledgements}
\begin{thebibliography}{}
\bibitem{Tomonaga50} S.\ Tomonaga, Prog.\ Theo.\ Phys.\ {\bf 5}, 544 (1950).
\bibitem{Luttinger63} J.\ M.\ Luttinger, J.\ Math.\ Phys.\ {\bf 4}, 1154 (1963).
\bibitem{Solyom79} J.\ S\'{o}lyom, Adv.\ Phys.\ {\bf 28}, 201 (1979).
\bibitem{Haldane81} F.\ D.\ M.\ Haldane, J.\ Phys.\ C {\bf 14}, 2585 (1981).
\bibitem{Voit95} J.\ Voit, Rep.\ Prog.\ Phys.\ {\bf 58},
977 (1995).
\bibitem{Giamarchi03}
T. Giamarchi, {\em Quantum Physics in One Dimension} (New York: Oxford
University Press, 2003).
\bibitem{Schoenhammer05}
K. Sch\"onhammer in
{\em Interacting Electrons in Low Dimensions} ed. by D. Baeriswyl (Dordrecht:
Kluwer Academic Publishers, 2005); arXiv:cond-mat/0305035.
\bibitem{vonDelft98}
J. von Delft and H. Schoeller, Annalen der Physik {\bf 7}, 225 (1998).
\bibitem{Meden99}
V.~Meden, Phys.~Rev.~B {\bf 60}, 4571 (1999).
\bibitem{Luther74} A.\ Luther and I.\ Peschel, Phys.\ Rev.\ B
{\bf 9}, 2911 (1974).
\bibitem{Dzyaloshinskii73} I.\ E.\ Dzyaloshinski\v{i} and A.\ I.\ Larkin,
Zh.\ Eksp.\ Teor.\ Fiz.\ {\bf 65}, 411 (1973)
[Sov.\ Phys.\ JETP {\bf 38}, 202 (1974)].
\bibitem{Theumann67} A.\ Theumann, J.\ Math.\ Phys.\ {\bf 8}, 2460
(1967).
\bibitem{Meden92}
V.\ Meden and K.\ Sch\"onhammer, Phys.\ Rev.\ B {\bf
46}, 15753 (1992).
\bibitem{Voit93}
J.\ Voit, Phys.\ Rev.\ B {\bf 47}, 6740 (1993).
\bibitem{Schoenhammer93} K.\ Sch\"onhammer and V.\ Meden, Phys.\ Rev.\ B
{\bf 47}, 16205 (1993).
\bibitem{Essler05}
F.H.L. Essler, H. Frahm, F. G\"ohmann, A. Kl\"umper, and V.E. Korepin,
{\em The One-Dimensional Hubbard Model} (Cambridge University Press, 2005).
\bibitem{Khodas07} M. Khodas, M. Pustilnik, A. Kamenev, and L.I. Glazman, Phys. Rev. B {\bf 76},
155402 (2007).
\bibitem{Pereira09} R.G. Pereira, S.R. White, and I. Affleck, Phys. Rev. B {\bf 79}, 165113 (2009).
\bibitem{Imambekov09a} A. Imambekov and L.I. Glazman, Science {\bf 323},228 (2009).
\bibitem{Imambekov09b} A. Imambekov and L.I. Glazman, Phys. Rev. Lett. {\bf 102}, 126405 (2009).
\bibitem{Schmidt10} T.L. Schmidt, A. Imambekov, and L.I. Glazman, Phys. Rev. Lett. {\bf 104}, 116403
(2010).
\bibitem{Essler15} F.H.L. Essler, R.G. Pereira, and I. Schneider, Phys. Rev. B {\bf 91}, 245150 (2015).
\bibitem{Imambekov12}
A. Imambekov, T.L. Schmidt, and L.I. Glazman, Rev. Mod. Phys. {\bf 84}, 1253 (2012).
\bibitem{Grioni09}
For a review on the experimental status until 2009 see M. Grioni, S. Pons and E. Frantzeskakis,
J.~Phys.: Condens.~Matter {\bf 21}, 023201 (2009).
\bibitem{Bohr81} T. Bohr in
{\em Low-dimensional Statistical Mechanics,} ed. by T. Bohr, A. Luther, and P. Minnhagen
(NORDITA preprint, ISSN 0106-2646, 1981).
\bibitem{Suzumura80}
Y. Suzumura, Prog. Theor. Phys. {\bf 63}, 51 (1980).
\bibitem{Schoenhammer93b}
K. Sch\"onhammer and V. Meden, Phys. Rev. B {\bf 48}, 11390 (1993).
\bibitem{Voit93b}
J. Voit, J. Phys.: Condens. Matter {\bf 5}, 8305 (1993).
\bibitem{Maebashi14}
H. Maebashi and Y. Takada, Phys. Rev. B {\bf 89}, 201109(R) (2014).
\bibitem{Markhof15} L. Markhof, Master thesis, RWTH Aachen University (2015).
\bibitem{Gebhard03}
F. Gebhard, E. Jeckelmann, S. Mahlert, S. Nishimoto, and R. M. Noack,
Eur. Phys. J. B {\bf 36}, 491 (2003).
\bibitem{Raas05}
C. Raas and G.S. Uhrig, Eur. Phys. J. B {\bf 45}, 293 (2005).
\bibitem{Senechal00}
D. S\'{e}n\'{e}chal, D. Perez, and M. Pioro-Ladri\`{e}re, Phys. Rev. Lett. {\bf 84},
522 (2000).
\bibitem{Zacher98}
M.G. Zacher, E. Arrigoni, W. Hanke, and J.R. Schrieffer,
Phys. Rev. B {\bf 57}, 6370 (1998).
\bibitem{Abendschein06}
A. Abendschein and F.F. Assaad, Phys. Rev. B {\bf 73}, 165119 (2006).
\bibitem{Benthien04} H. Benthien, F. Gebhard, and E. Jeckelmann, Phys. Rev. Lett. {\bf 92}, 256401
(2004).
\bibitem{Carmelo06} J.M.P. Carmelo, K. Penc, P.D. Sacramento, M. Sing, and R. Claessen,
J. Phys.: Condens. Matter {\bf 18}, 5191 (2006).
\bibitem{Ohtsubo15}
Y. Ohtsubo, J. Kishi, K. Hagiwara, P. Le F\`{e}vre, F. Bertran, A. Taleb-Ibrahimi,
M. Matsunami, K. Tanaka, and S. Kimura, arXiv:1504.04081v1 (2015).
\bibitem{Schoenhammer93c}
K. Sch\"onhammer and V. Meden, J. Electr.~Spectrosc.~Relat.~Phenom. {\bf 62}, 225 (1993).
\bibitem{Nakamura97}
N. Nakamura and Y. Suzumura, Prog. Theor. Phys. {\bf 98}, 29 (1997).
\bibitem{Orgad01}
D. Orgad, Philos. Mag. B {\bf 81}, 375 (2001).
\end{thebibliography}{}
\end{document}