Matsubara Formalism Preview

Matsubara formalism is equilibrium quantum mechanics reorganized on a compact imaginary-time interval. It represents thermal correlators by discrete imaginary-frequency samples. Imaginary-time convolutions become products, while products of time-dependent factors generate frequency sums. Perturbative many-body calculations can then be organized inside a thermal trace.

The essential choices are the equilibrium generator, the operator channel, and its parity. Once those are fixed, thermal boundary conditions determine the frequency grid, and a Fourier series organizes exact or approximate calculations. Real-frequency response requires a further analytic continuation.

Helpful background. Imaginary time defines the thermal interval and operator evolution; thermal density operators explains equilibrium trace averages.

Thermal generators and Matsubara conventions

Equilibrium generator

Let

$$\mathcal K = H-\mu N,$$

with $[H,N]=0$ when a chemical potential is present. The grand partition function and equilibrium average are

$$\begin{aligned} \Xi &= \operatorname{Tr} e^{-\beta\mathcal K}, \\ \langle O\rangle_\beta &= \frac1\Xi \operatorname{Tr} \left( e^{-\beta\mathcal K}O \right). \end{aligned}$$

Assume $\beta>0$ and a finite Gibbs trace, or a specified thermodynamic limit of local correlators. For a canonical ensemble, set $\mathcal K=H$ and write $Z$ instead of $\Xi$. A fixed-$N$ trace permits the following simple cyclic argument for sector-preserving operators. The single-particle fields and Wick expansion below instead use the full grand-canonical Fock-space trace; a number projection must be treated explicitly.

Imaginary-time operators evolve with the same generator used in the density operator:

$$O(\tau) = e^{\tau\mathcal K/\hbar} O e^{-\tau\mathcal K/\hbar}.$$

Evolving with $H$ while weighting states with $H-\mu N$ mixes energy origins. One can translate between conventions, but one cannot combine them silently.

Thermal interval

Define the thermal circumference

$$L_\tau := \beta\hbar, \qquad \beta = \frac1{k_{\mathrm B}T}.$$

The basic interval is

$$0 \leq \tau < L_\tau.$$

The endpoint is identified with the beginning in the thermal trace. A sign appears in the boundary condition of an ordered correlator when odd operators are exchanged; ordinary trace cyclicity itself is sign-free.

Energy and angular frequency

This page uses Matsubara energies

$$\nu_n = \hbar\omega_n.$$

Thus $\nu_n$ has energy units, while $\omega_n$ has inverse-time units. The Fourier phase is

$$e^{-i\nu_n\tau/\hbar} = e^{-i\omega_n\tau}.$$

Many texts set $\hbar=k_{\mathrm B}=1$ and call the energy variable $\omega_n$. Restoring units requires identifying which quantity the symbol represented.

Fermion parity

Assign a parity

$$p_O \in \{0,1\},$$

where $p_O=0$ for an even operator and $p_O=1$ for an odd operator. Under fermion-parity conjugation,

$$(-1)^{N_{\mathrm F}} O(-1)^{N_{\mathrm F}} = (-1)^{p_O}O.$$

Here $N_{\mathrm F}$ counts fermions, rather than a possible boson number used in the chemical potential. Typical examples are:

Operator or channel	Parity	Thermal class
$c$, $c^\dagger$	odd	fermionic
$\phi$, $b$, $b^\dagger$	even	bosonic
density $c^\dagger c$	even	bosonic
spin or current observable	even	bosonic
pair field $c_\downarrow c_\uparrow$	even	bosonic
product of one fermion and an even observable	odd	fermionic

The statistics of the constituents do not by themselves determine the frequency grid. The parity of the full operator channel does.

From the Trace to Compact Imaginary Time

The formal resemblance

$$e^{-\beta\mathcal K} = e^{-L_\tau\mathcal K/\hbar}$$

turns the Gibbs operator into evolution through an imaginary-time interval of length $L_\tau$. Taking the trace closes the interval.

For $0<\tau<L_\tau$, define an unordered thermal correlator

$$C_{AB}^{>}(\tau) := \left\langle A(\tau)B(0) \right\rangle_\beta.$$

Trace cyclicity gives

$$C_{AB}^{>}(\tau) = \left\langle B(0)A(\tau-L_\tau) \right\rangle_\beta.$$

This is the operator core of the Kubo–Martin–Schwinger relation. Trace cyclicity itself introduces no fermionic minus sign. The minus sign appears when odd operators are exchanged in the graded time-ordered correlator used to glue the two branches into one function on the thermal circle.

KMS Condition derives this equilibrium boundary condition and its spectral consequences.

Graded Imaginary-Time Ordering

For operators of definite parity, imaginary-time ordering is

$$\mathcal T_\tau A(\tau)B(\tau') = \theta(\tau-\tau') \times A(\tau)B(\tau') + (-1)^{p_Ap_B} \theta(\tau'-\tau) \times B(\tau')A(\tau).$$

When both operators are odd, exchanging them contributes a minus sign. Otherwise the exchange is even.

A channel-dependent correlator can then be written schematically as

$$\mathcal C_{AB}(\tau-\tau') = s_{AB} \left\langle \mathcal T_\tau A(\tau)B(\tau') \right\rangle_\beta,$$

where the overall sign $s_{AB}$ is a convention. For example, the normal fermionic single-particle Green function commonly uses $s_{AB}=-1$. The boundary class does not depend on that fixed overall sign.

Equal-time definitions require a one-sided limit. For canonical fields, the difference between $0^+$ and $0^-$ contains the commutator or anticommutator contact term. Replacing both by an undefined value at $\tau=0$ loses occupation and normalization information.

Thermal Boundary Conditions

For a two-point channel generated by an operator of parity $p$, the ordered correlator obeys

$$\mathcal C_p(\tau+L_\tau) = (-1)^p \mathcal C_p(\tau).$$

Therefore:

$$\begin{aligned} \mathcal C_{\mathrm B}(\tau+L_\tau) &= \mathcal C_{\mathrm B}(\tau), \\ \mathcal C_{\mathrm F}(\tau+L_\tau) &= - \mathcal C_{\mathrm F}(\tau). \end{aligned}$$

These statements concern thermal correlators or coherent-state fields, not the literal sign of a physical many-body state after elapsed clock time.

Composite operators

Two fermionic factors make an even operator. Thus

$$n_i = c_i^\dagger c_i$$

has bosonic boundary conditions. So does a pair field

$$\Delta_i = c_{i\downarrow}c_{i\uparrow}.$$

By contrast, a normal single-particle function built from $c_i(\tau)$ is fermionic. Confusing constituent statistics with channel parity assigns the wrong zero mode and the wrong frequency grid.

Symmetry qualification

If the equilibrium state preserves fermion parity, an expectation value with odd total parity vanishes. Thermal Green functions remain nonzero because their complete operator product is even, even though shifting one odd insertion around the circle produces antiperiodicity.

Discrete Matsubara Energies

A mode on the thermal circle has the form

$$e^{-i\nu\tau/\hbar}.$$

The boundary condition requires

$$e^{-i\beta\nu} = (-1)^p.$$

Hence the compact formula for a parity-$p$ channel is

$$\nu_n^{(p)} = \frac{(2n+p)\pi}{\beta}, \qquad n\in\mathbb Z.$$

The two grids are

$$\begin{aligned} \Omega_m &= \frac{2\pi m}{\beta} && \text{bosonic}, \\ \nu_n &= \frac{(2n+1)\pi}{\beta} && \text{fermionic}. \end{aligned}$$

Here $\Omega_m$ labels a bosonic external or internal energy and $\nu_n$ a fermionic one. This separate notation reduces routing mistakes.

The corresponding angular frequencies are

$$\omega_n^{(p)} = \frac{\nu_n^{(p)}}{\hbar} = \frac{(2n+p)\pi}{\beta\hbar}.$$

The bosonic grid contains $\Omega_0=0$. The fermionic grid is shifted by half a spacing and has no zero mode.

Fourier Transform on the Thermal Circle

With the energy convention above, use the transform pair

$$\mathcal C(i\nu_n) = \frac1\hbar \int_0^{L_\tau} d\tau\, e^{i\nu_n\tau/\hbar} \mathcal C(\tau),$$

and

$$\mathcal C(\tau) = \frac1\beta \sum_{n\in\mathbb Z} e^{-i\nu_n\tau/\hbar} \mathcal C(i\nu_n).$$

The grid is bosonic or fermionic according to the channel. The normalization follows from

$$\frac1{L_\tau} \int_0^{L_\tau} d\tau\, e^{i(\nu_n-\nu_m)\tau/\hbar} = \delta_{nm}.$$

Equivalently, the twisted delta distribution is

$$\delta_p(\tau-\tau') = \frac1{\beta\hbar} \sum_n e^{-i\nu_n^{(p)}(\tau-\tau')/\hbar}.$$

For $p=0$ it is periodic. For $p=1$ it changes sign when either argument is shifted by $L_\tau$.

Dimensional check

Because $d\tau/\hbar$ has inverse-energy units, if $\mathcal C(\tau)$ is dimensionless then $\mathcal C(i\nu_n)$ has inverse-energy units. The inverse transform carries $1/\beta$, which has energy units. This check catches many missing factors.

Frequency Arithmetic

The two grids close under the combinations required at interaction vertices:

$$\begin{aligned} \Omega_m+\Omega_\ell &= \Omega_{m+\ell}, \\ \nu_n+\Omega_m &= \nu_{n+m}, \\ \nu_n-\nu_r &= \Omega_{n-r}. \end{aligned}$$

Thus:

• boson plus boson is bosonic;
• fermion plus boson is fermionic;
• fermion minus fermion is bosonic.

This arithmetic is the frequency-space expression of fermion-parity conservation.

For example, a two-line object with an external bosonic energy can have the schematic form

$$\mathcal I(i\Omega_m) = \frac1\beta \sum_n F(i\nu_n) G(i\nu_n+i\Omega_m).$$

Both internal arguments are fermionic because adding a bosonic grid point preserves the fermionic grid.

Products and convolutions in frequency space

For two correlators on compatible thermal grids, a product in time generates a convolution of frequencies. Conversely, an integral convolution in time generates a product in frequency. For example, with the $1/\hbar$ convolution normalization,

$$C(\tau)=\frac1\hbar\int_0^{\beta\hbar}d\tau'\, A(\tau-\tau')B(\tau') \quad\Longrightarrow\quad C(i\nu_n)=A(i\nu_n)B(i\nu_n).$$

The functions must be extended with the same periodic or antiperiodic boundary condition. This is why Dyson equations and linear integral equations become algebraic at each Matsubara frequency, while interactions still generate sums over internal frequencies.

Imaginary-Time Derivatives

Away from contact points,

$$\partial_\tau e^{-i\nu_n\tau/\hbar} = -\frac{i\nu_n}{\hbar} e^{-i\nu_n\tau/\hbar}.$$

Therefore a first-order imaginary-time equation becomes algebraic in frequency space. Schematically,

$$\left( \hbar\partial_\tau+\xi \right) \mathcal G_0(\tau) = -\hbar\delta_{\mathrm F}(\tau)$$

transforms into

$$\left( -i\nu_n+\xi \right) \mathcal G_0(i\nu_n) = -1.$$

Hence

$$\mathcal G_0(i\nu_n) = \frac1{i\nu_n-\xi}.$$

The delta function records the equal-time anticommutator. Dropping the contact term gives the homogeneous equation and cannot determine the Green function’s normalization.

The exact sign on the right-hand side follows from the chosen Green-function convention. Green Functions in Many-Body QM fixes that convention and derives the free benchmark.

Interaction Expansion on the Thermal Circle

Split the ensemble generator as

$$\mathcal K = \mathcal K_0+V.$$

In the interaction picture defined by $\mathcal K_0$,

$$V_I(\tau) = e^{\tau\mathcal K_0/\hbar} V e^{-\tau\mathcal K_0/\hbar}.$$

The partition-function ratio is

$$\frac{\Xi}{\Xi_0} = \left\langle \mathcal T_\tau \exp\left[ -\frac1\hbar \int_0^{L_\tau} d\tau\, V_I(\tau) \right] \right\rangle_0.$$

An interacting ordered correlator has the normalized form

$$\frac{ \left\langle \mathcal T_\tau \mathcal O e^{-\hbar^{-1}\int d\tau\,V_I} \right\rangle_0 }{ \left\langle \mathcal T_\tau e^{-\hbar^{-1}\int d\tau\,V_I} \right\rangle_0 }.$$

The denominator cancels disconnected vacuum contributions. If the reference Gibbs state is Gaussian, Wick’s theorem reduces products of linear fields to pair contractions (and first moments if bosonic fields are displaced). A stable quadratic generator traced over full Fock space gives such a reference. A fixed-number projection is generally non-Gaussian even for a quadratic Hamiltonian; its higher correlations need separate treatment.

Wick’s theorem develops contraction algebra; diagrammatic methods organizes the resulting graphs and self-energies.

Why Frequency Sums Appear

Each internal imaginary-time variable is integrated over a compact interval. Expanding every line in its Fourier series converts a vertex integral into a Kronecker constraint among discrete frequencies. Independent closed frequency routes remain summed.

In a translation-invariant continuum system, a typical loop measure is

$$\frac1\beta \sum_n \int \frac{d^d k}{(2\pi)^d}.$$

For lattice models, the momentum integral is replaced by the appropriate Brillouin-zone sum or finite-size momentum grid.

A common schematic structure is obtained by abbreviating

$$\int_{\mathbf q} := \int \frac{d^d q}{(2\pi)^d}.$$

Then

$$\Sigma(\mathbf k,i\nu_n) = \frac1\beta \sum_m \int_{\mathbf q} \times \mathcal V\, D(\mathbf q,i\Omega_m) \times G(\mathbf k-\mathbf q,i\nu_n-i\Omega_m).$$

The displayed expression communicates routing only. Vertex tensors, signs, symmetry factors, matrix indices, and powers of $\hbar$ depend on the Hamiltonian and the propagator convention.

Worked Sum: A Bosonic Thermal Denominator

Let $E>0$ and consider the absolutely convergent bosonic sum

$$S_{\mathrm B}(E) := \frac1\beta \sum_{m=-\infty}^{\infty} \frac1{\Omega_m^2+E^2},$$

where

$$\Omega_m = \frac{2\pi m}{\beta}.$$

Set

$$a := \frac{\beta E}{2\pi}.$$

Then

$$S_{\mathrm B}(E) = \frac{\beta}{4\pi^2} \sum_{m=-\infty}^{\infty} \frac1{m^2+a^2}.$$

Using

$$\sum_{m=-\infty}^{\infty} \frac1{m^2+a^2} = \frac{\pi}{a} \coth(\pi a),$$

one obtains

$$S_{\mathrm B}(E) = \frac1{2E} \coth\left( \frac{\beta E}{2} \right).$$

In terms of the Bose occupation

$$n_{\mathrm B}(E) = \frac1{e^{\beta E}-1},$$

the same result is

$$S_{\mathrm B}(E) = \frac{ 1+2n_{\mathrm B}(E) }{ 2E }.$$

This identity separates a zero-temperature contribution from thermal population.

Limiting checks

At zero temperature,

$$\lim_{\beta\to\infty} S_{\mathrm B}(E) = \frac1{2E}.$$

For $\beta E\ll1$,

$$S_{\mathrm B}(E) = \frac1{\beta E^2} + \frac{\beta}{12} + O(\beta^3E^2).$$

The leading term is exactly the $m=0$ contribution. This is the simplest demonstration of why a bosonic zero mode can dominate a high-temperature or long-distance observable.

Worked sum: two fermionic denominators

Let $a,b$ be real one-particle energies and $\Omega_m$ a bosonic Matsubara energy. Partial fractions give

$$\begin{aligned} I_m(a,b)&=\frac1\beta\sum_n \frac1{(i\nu_n-a)(i\nu_n+i\Omega_m-b)}\\ &=\frac{f(a)-f(b)}{a-b+i\Omega_m}. \end{aligned}$$

The original sum is absolutely convergent. The two single-denominator sums after partial fractions each need a common prescription, explained next; their equal-time constants cancel. Shifting a fermionic index by the integer $m$ leaves its infinite grid unchanged.

For $a=b$ and $m=0$, take the limit of the quotient:

$$I_0(a,a)=f'(a)=-\beta f(a)[1-f(a)].$$

For $a=b$ and $m\ne0$, the sum is zero. Thus the connected number correlator of one isolated free fermionic mode is

$$C_{nn}(i\Omega_m)=-I_m(a,a) =\beta f(a)[1-f(a)]\delta_{m0}.$$

The number is conserved, so its fluctuation is constant in imaginary time and appears entirely in the zero Matsubara component. Its isolated retarded commutator response vanishes. This is a concrete reason to keep static contributions separate when continuing a bosonic correlator.

Convergence Prescriptions and Equal-Time Limits

Not every Matsubara sum is absolutely convergent. The free fermionic propagator behaves as

$$\mathcal G_0(i\nu_n) \sim \frac1{i\nu_n}$$

at large $\lvert n\rvert$. Its equal-time inverse transform must retain the side from which $\tau=0$ is approached.

For the normal convention

$$\mathcal G_0(i\nu_n) = \frac1{i\nu_n-\xi},$$

one has

$$\frac1\beta \sum_n \frac{ e^{+i\nu_n0^+/\hbar} }{ i\nu_n-\xi } = f(\xi),$$

where

$$f(\xi) = \frac1{e^{\beta\xi}+1}.$$

The opposite side gives

$$\frac1\beta \sum_n \frac{ e^{-i\nu_n0^+/\hbar} }{ i\nu_n-\xi } = - \left[ 1-f(\xi) \right].$$

Their difference is the canonical equal-time discontinuity:

$$\mathcal G_0(0^-) - \mathcal G_0(0^+) = 1.$$

Writing an unqualified sum of $1/(i\nu_n-\xi)$ and discarding the convergence factor hides which one-sided value is intended.

Static Sector and Bosonic Zero Modes

A static equilibrium perturbation carries zero bosonic Matsubara energy:

$$\Omega_0 = 0.$$

That does not mean every zero-frequency limit is interchangeable. For a response function $\chi(\mathbf q,i\Omega_m)$, the limits

$$\lim_{\mathbf q\to0} \chi(\mathbf q,i\Omega_0)$$

and

$$\lim_{E\to0} \chi^{\mathrm R}(\mathbf 0,E)$$

can represent different physical protocols. Conservation laws, hydrodynamic poles, and symmetry breaking make the order of limits important.

Fermionic single-particle lines have no zero Matsubara mode. Composite fermion bilinears are even, however, and can carry zero bosonic external frequency.

Susceptibilities explains the static and dynamical response limits in more detail.

Zero-Temperature Limit

As $\beta\to\infty$, the spacing of either Matsubara grid tends to zero:

$$\Delta\nu = \frac{2\pi}{\beta} \longrightarrow 0.$$

For a sufficiently regular and decaying integrand,

$$\frac1\beta \sum_n F(i\nu_n) \longrightarrow \int_{-\infty}^{\infty} \frac{d\nu}{2\pi} F(i\nu).$$

In angular-frequency variables, the corresponding statement is

$$\frac1{\beta\hbar} \sum_n F(i\omega_n) \longrightarrow \int_{-\infty}^{\infty} \frac{d\omega}{2\pi} F(i\omega).$$

The limit is not automatic near infrared singularities, a closing gap, a Fermi surface, or a phase transition. One must control convergence and the order of the thermodynamic, zero-temperature, and zero-frequency limits.

Spectral and Real-Time Bridge

For many standard channels, Matsubara values are samples of an analytic function

$$\mathcal G(z) = \int_{-\infty}^{\infty} dE\, \frac{\rho(E)}{z-E},$$

evaluated at

$$z = i\nu_n.$$

For the convention $G=-\langle\mathcal T_\tau AB\rangle$, this is the dynamical Cauchy transform with the corresponding commutator or anticommutator spectral density. A bosonic static term proportional to $\delta_{m0}$ is additional data, as in the number example above. Physical response defined with $+i\theta[A(t),B]/\hbar$ for a source $-fB$ requires the corresponding opposite sign from the $-i$ retarded Green-function convention.

The retarded function is obtained from the upper half-plane, containing the positive Matsubara frequencies:

$$\mathcal G^{\mathrm R}(E) = \lim_{\epsilon\downarrow0} \mathcal G(E+i\epsilon).$$

The mnemonic

$$i\nu_n \longrightarrow E+i0^+$$

is valid only after the analytic function and its convention have been identified. It is not a substitution rule for a finite list of noisy values.

Exact discrete data together with appropriate analyticity and asymptotic conditions can determine the continuation in principle. Numerical reconstruction from finite uncertain data is ill posed. Sum rules, uncertainties, priors, and resolution tests are part of the result. Analytic Continuation develops that inverse problem.

Thermal Green Functions derives the single-particle spectral relation; Spectral Functions explains how to interpret reconstructed peaks and continua.

From Finite Systems to Thermal Field Theory

For finitely many orbitals, the Matsubara formalism is already a complete operator method. Moving to field theory adds:

• a spatial coordinate or momentum label;
• infinitely many degrees of freedom;
• local interactions and composite operators;
• ultraviolet regularization and renormalization;
• gauge constraints when gauge fields are present;
• infrared questions associated with zero modes and collective behavior.

The thermal circle and frequency grids survive unchanged in principle. A free inverse propagator becomes a function such as

$$\mathcal G_0^{-1} (\mathbf k,i\nu_n),$$

and loop calculations use both momentum integrals and Matsubara sums.

For a relativistic or linearly dispersing mode of speed $v$, distances much longer than the thermal length $\beta\hbar v$ can make nonzero temporal modes irrelevant to static long-distance behavior. A quadratic dispersion has a different thermal length, of order $\hbar\sqrt{\beta/m}$; there is no universal conversion from imaginary-time extent to distance. A bosonic zero mode may then support an effective lower-dimensional statistical field theory. Fermions lack a zero mode, although they still affect matching coefficients and interactions. Scale separation, infrared dynamics, and gauge structure decide whether this dimensional-reduction picture is controlled.

From Euclidean Time to Euclidean QFT gives the internal bridge. Full relativistic thermal field theory, renormalization, gauge holonomy, and real-time thermal contours belong in the corresponding QFT treatment.

Representation Guide

Goal	Natural representation	Main caution
free energy or equilibrium derivative	trace, $\ln\Xi$, linked diagrams	normalize and remove disconnected pieces
imaginary-time decay or gap estimate	$\mathcal C(\tau)$	finite-temperature wrap-around terms
perturbative equilibrium correction	$\mathcal G(i\nu_n)$	signs, routing, and sum convergence
static response	bosonic $\Omega_0$ sector	order of momentum and frequency limits
real-frequency spectrum	spectral or retarded function	continuation is not direct substitution
strongly coupled Euclidean observable	path integral or thermal-state method	discretization, finite size, and sign problem
explicit nonequilibrium evolution	real-time or contour method	Matsubara equilibrium data are insufficient

Matsubara methods are especially effective when equilibrium, translation invariance, and a controlled Gaussian or perturbative reference are available. They are not a universal replacement for spectral diagonalization, Monte Carlo, tensor networks, or real-time evolution.

Common pitfalls

Shifting a cutoff sum can change its boundary terms. Frequency routing is exact on the infinite grid. A truncated or conditionally convergent sum requires the regulator and omitted tail to be carried through the shift.

A zero Matsubara component is not always a retarded limit. Conserved fluctuations can give a static contribution with no isolated commutator response. Decide whether the desired measurement permits re-equilibration before taking a zero-frequency limit.

A diagram does not determine its own sign. Field definitions, closed fermion loops, vertex conventions, and symmetry factors must come from one consistent expansion. A free-mode or atomic calculation is a useful independent check.

Continue with the thermal calculation

• Matsubara frequency table collects units, indexing, parity, and cutoff conventions.
• Thermal Green functions derives endpoint values and exact free-mode propagators.
• Spectral representation connects Lehmann sums to Euclidean and retarded functions.
• Finite-temperature QFT develops thermal propagators, loop integrals, screening, and dimensional reduction.

Exercises

1. Classify thermal channels

For a fermionic lattice model, classify the following as bosonic or fermionic Matsubara channels:

$$\begin{aligned} &c_i, \qquad n_i=c_i^\dagger c_i, \\ &\Delta_i = c_{i\downarrow}c_{i\uparrow}, \qquad c_i n_j. \end{aligned}$$

State whether a zero Matsubara mode is allowed.

Solution

The operator $c_i$ is odd, so its channel is fermionic and antiperiodic. It has no zero Matsubara mode.

The density contains two fermionic factors:

$$p_{n_i} = 1+1 \pmod 2 = 0.$$

It is bosonic and can have a zero mode. The pair field also contains two odd factors and is therefore bosonic:

$$p_{\Delta_i} = 0.$$

The product $c_i n_j$ is odd because an odd operator times an even operator remains odd. Its channel is fermionic and has no zero mode.

2. Derive the two grids

Insert the mode

$$f_\nu(\tau) = e^{-i\nu\tau/\hbar}$$

into periodic and antiperiodic boundary conditions on an interval of length $\beta\hbar$. Derive the allowed Matsubara energies.

Solution

Periodicity requires

$$e^{-i\beta\nu} = 1.$$

Therefore

$$\beta\nu = 2\pi m,$$

and

$$\Omega_m = \frac{2\pi m}{\beta}.$$

Antiperiodicity requires

$$e^{-i\beta\nu} = -1,$$

so

$$\beta\nu = (2n+1)\pi.$$

Hence

$$\nu_n = \frac{(2n+1)\pi}{\beta}.$$

Only the periodic grid contains zero.

3. Check the transform normalization

Starting from

$$\mathcal C(i\nu_n) = \frac1\hbar \int_0^{\beta\hbar} d\tau\, e^{i\nu_n\tau/\hbar} \mathcal C(\tau),$$

derive the inverse coefficient $1/\beta$.

Solution

Expand

$$\mathcal C(\tau) = \sum_n a_n e^{-i\nu_n\tau/\hbar}.$$

Multiply by $e^{i\nu_m\tau/\hbar}$ and integrate:

$$\int_0^{\beta\hbar} d\tau\, e^{i(\nu_m-\nu_n)\tau/\hbar} = \beta\hbar\, \delta_{mn}.$$

Thus

$$\mathcal C(i\nu_m) = \frac1\hbar \left( \beta\hbar \right) a_m = \beta a_m.$$

Therefore

$$a_m = \frac1\beta \mathcal C(i\nu_m),$$

which gives the stated inverse transform.

4. Fermionic partner of the bosonic sum

Use

$$\sum_{n=-\infty}^{\infty} \frac1{(n+\tfrac12)^2+a^2} = \frac{\pi}{a} \tanh(\pi a)$$

to evaluate

$$S_{\mathrm F}(E) = \frac1\beta \sum_n \frac1{\nu_n^2+E^2}, \qquad E>0.$$

Express the result using the Fermi function.

Solution

With

$$\nu_n = \frac{2\pi}{\beta} \left( n+\frac12 \right),$$

and $a=\beta E/(2\pi)$,

$$S_{\mathrm F}(E) = \frac{\beta}{4\pi^2} \sum_n \frac1{(n+\tfrac12)^2+a^2}.$$

The given identity yields

$$S_{\mathrm F}(E) = \frac1{2E} \tanh\left( \frac{\beta E}{2} \right).$$

Since

$$1-2f(E) = \tanh\left( \frac{\beta E}{2} \right),$$

the result is

$$S_{\mathrm F}(E) = \frac{ 1-2f(E) }{ 2E }.$$

Unlike the bosonic sum, it has no isolated zero-mode contribution.

5. Route a mixed loop

An external fermionic line carries $\nu_n$, and an internal bosonic line carries $\Omega_m$. Show that the remaining internal fermionic line may carry $\nu_n-\Omega_m$. Find its integer label.

Solution

Using the definitions,

$$\begin{aligned} \nu_n-\Omega_m &= \frac{(2n+1)\pi}{\beta} - \frac{2m\pi}{\beta} \\ &= \frac{ \left[ 2(n-m)+1 \right]\pi }{ \beta }. \end{aligned}$$

This is the fermionic frequency

$$\nu_{n-m}.$$

Subtracting a bosonic grid point therefore preserves the fermionic grid, as required by parity conservation.

6. Recover the zero-temperature measure

Let $F(i\nu)$ be smooth and decay fast enough at large $\lvert\nu\rvert$. Show heuristically that

$$\frac1\beta \sum_n F(i\nu_n)$$

approaches an integral as $\beta\to\infty$, for either parity grid.

Solution

Adjacent points on either grid are separated by

$$\Delta\nu = \frac{2\pi}{\beta}.$$

Therefore

$$\frac1\beta = \frac{\Delta\nu}{2\pi}.$$

The sum is a Riemann sum:

$$\frac1\beta \sum_n F(i\nu_n) = \sum_n \frac{\Delta\nu}{2\pi} F(i\nu_n).$$

As $\beta\to\infty$, $\Delta\nu\to0$, so

$$\frac1\beta \sum_n F(i\nu_n) \longrightarrow \int_{-\infty}^{\infty} \frac{d\nu}{2\pi} F(i\nu).$$

The half-step shift between the bosonic and fermionic grids vanishes with the spacing. Smoothness, decay, and uniform control are needed; infrared singularities or a simultaneous thermodynamic limit can invalidate the naive interchange of limits.

7. Diagnose an invalid continuation

A numerical calculation returns $\mathcal G(i\nu_n)$ at the lowest 20 fermionic Matsubara energies with error bars. A fit replaces $i\nu_n$ by $E+i0^+$ directly in those 20 values and reports three sharp spectral peaks. Identify the conceptual failure and state what checks are needed.

Solution

The values form a finite discrete data set, not an analytic expression. Replacing their arguments does not construct the analytic function whose imaginary-axis samples they approximate.

A defensible continuation must specify a spectral or analytic model, propagate the data covariance, enforce known asymptotics and sum rules, and test resolution with synthetic data or controlled benchmarks. Stability against changing the frequency window, regularization, and prior should be reported. Three peaks are justified only if alternative spectra compatible with the uncertainties cannot erase or merge them at the claimed resolution.

References

• A. A. Abrikosov, L. P. Gorkov, and I. E. Dzyaloshinski, Methods of Quantum Field Theory in Statistical Physics, Dover (1975) – classic diagrammatic finite-temperature methods.
• G. Baym and N. D. Mermin, “Determination of Thermodynamic Green’s Functions”, Journal of Mathematical Physics 2, 232–234 (1961) – conditions connecting discrete thermal data to analytic Green functions.
• A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems, Dover (2003 reprint of the 1971 edition) – standard operator and finite-temperature Green-function treatment.
• R. Kubo, “Statistical-Mechanical Theory of Irreversible Processes. I. General Theory and Simple Applications to Magnetic and Conduction Problems”, Journal of the Physical Society of Japan 12, 570–586 (1957) – equilibrium correlation and response foundations.
• M. Le Bellac, Thermal Field Theory, Cambridge University Press (1996) – bridge to relativistic finite-temperature field theory.
• G. D. Mahan, Many-Particle Physics, 3rd ed., Springer (2000) – Matsubara sums, response, and condensed-matter applications.
• P. C. Martin and J. Schwinger, “Theory of Many-Particle Systems. I”, Physical Review 115, 1342–1373 (1959) – thermal Green functions and equilibrium many-particle identities.
• T. Matsubara, “A New Approach to Quantum-Statistical Mechanics”, Progress of Theoretical Physics 14, 351–378 (1955) – original imaginary-time many-body formalism.
• J. W. Negele and H. Orland, Quantum Many-Particle Systems, CRC Press (2018 reissue) – operator, coherent-state, and functional methods.