Thermal Strings and the Hagedorn Transition

An exponentially large number of massive string states can compete with Boltzmann suppression. The same effect appears in the Euclidean worldsheet description as a winding state becoming massless. Understanding why these descriptions agree requires keeping track of three different objects: a trace over one-string states, the partition function of a gas, and a connected torus amplitude.

We will derive this relation for weakly coupled type II strings in flat ten-dimensional spacetime, with the metric treated as a fixed background. The result is a perturbative Hagedorn threshold, not a derivation of the interacting high-temperature phase. The global spacetime and worldsheet conventions, the torus measure and spin structures, and the circle momentum and winding lattice supply the prerequisites.

Thermal traces and Euclidean winding

Let $Y=X_E^0$ have circumference $\beta$, so that $Y\sim Y+\beta$ and $R_0=\beta/(2\pi)$. The temperature is $T=\beta^{-1}$. For a Hamiltonian $H$, the ordinary thermal trace and free energy are

$$Z(\beta)=\operatorname{Tr}e^{-\beta H},\qquad W(\beta)=\log Z(\beta),\qquad F(\beta)=-\frac{W(\beta)}{\beta}.$$

The trace glues the final configuration to the initial one. Bosonic fields are periodic on the resulting Euclidean circle. A fermionic coherent-state trace contains an additional minus sign in this gluing, so spacetime fermions are antiperiodic. Consequently their frequencies are

$$\omega_n^{\mathrm B}=\frac{2\pi n}{\beta},\qquad \omega_n^{\mathrm F}=\frac{2\pi(n+1/2)}{\beta},\qquad n\in\mathbb Z.$$

Periodic spacetime fermions instead compute $\operatorname{Tr}(-1)^{\mathsf F}e^{-\beta H}$, where $(-1)^{\mathsf F}$ is spacetime fermion parity. This graded trace becomes a supersymmetric index only when supersymmetry, the spectrum, and convergence justify cancellation of paired states. It is not the ordinary thermal ensemble. Exercise 1 makes the distinction in a finite oscillator system.

Particles also have paths that wrap Euclidean time: these are worldline windings. Strings have the additional possibility that the spatial circle of the worldsheet wraps the thermal circle. In a torus path integral both worldsheet cycles can wind. With the coordinates used on the preceding page,

$$\zeta=u+\tau v,\qquad Y=\beta(bu+av),\qquad a,b\in\mathbb Z,$$

where $u$ and $v$ each have period one. The $A$ cycle changes $Y$ by $\beta b$, and the $B$ cycle changes it by $\beta a$. These integer map labels are distinct from the worldsheet spin structures.

The target circle has circumference $\beta$. The coordinate cuts of the worldsheet polygon carry $\Delta_A Y=\beta b$ and $\Delta_B Y=\beta a$; matching opposite edges are identified. The arrows specify orientation, and the drawing is schematic. Spacetime spin transport constrains the worldsheet spin sum but does not identify the two kinds of boundary condition.

From a one-string trace to a free gas

Let $z_{1\mathrm B}(\beta)$ and $z_{1\mathrm F}(\beta)$ be sums of $e^{-\beta E}$ over single-string bosonic and fermionic states, respectively. They include spatial momenta and physical polarization multiplicities. A bosonic mode with energy $E$ has occupation numbers $0,1,2,\ldots$, whereas a fermionic mode has occupations $0,1$. Their contributions to the gas partition function are $(1-e^{-\beta E})^{-1}$ and $1+e^{-\beta E}$. Taking the logarithm gives

$$W_{\mathrm{gas}}(\beta)=\sum_{r=1}^{\infty}\frac{z_{1\mathrm B}(r\beta)+(-1)^{r+1}z_{1\mathrm F}(r\beta)}{r}.$$

Thus $W_{\mathrm{gas}}$ is not exactly the one-string trace. Its $r=1$ term controls the leading high-energy exponential singularity, while all $r$ are needed for the Bose and Fermi occupation factors. A connected genus-one worldsheet computes $W$ at this order; exponentiating connected contributions produces $Z$.

For one bosonic species of mass $M$ in $d$ noncompact spatial dimensions, the vacuum-subtracted free energy is

$$\frac{F_{\mathrm B}}{V_d}=\frac1\beta\int\frac{d^d k}{(2\pi)^d}\log\bigl(1-e^{-\beta\sqrt{k^2+M^2}}\bigr).$$

Expanding the logarithm and applying the Gaussian identity

$$e^{-a\sqrt{b}}=\frac{a}{2\sqrt\pi}\int_0^\infty\frac{ds}{s^{3/2}}e^{-bs-a^2/(4s)},\qquad a,b>0,$$

turns the square-root Hamiltonian into a proper-time integral. The spatial momentum integral contributes $(4\pi s)^{-d/2}$. The factor $r\beta$ from the identity cancels $1/(r\beta)$ from the logarithm, giving

$$\frac{F_{\mathrm B}}{V_d}=-\frac1{(4\pi)^{(d+1)/2}}\sum_{r=1}^{\infty}\int_0^\infty\frac{ds}{s^{(d+3)/2}}e^{-sM^2-r^2\beta^2/(4s)}.$$

For a fermionic species, insert $(-1)^{r+1}$ in this sum. The proper time $s$ has dimensions of length squared. The factor $e^{-r^2\beta^2/(4s)}$ is the classical action of a worldline wrapping the thermal circle $r$ times. The omitted $r=0$ term is the temperature-independent vacuum contribution. Opposite orientations have already been combined in the positive-$r$ sum.

This particle calculation fixes the normalization with which the modular string expression must agree. The relation between the free gas, the strip integral, and the long-string limit is developed in Kruczenski and Lawrence, sections 2–3 and Mertens, Verschelde, and Zakharov, sections 2–3.

The thermal spin sum

Write a chiral worldsheet spin structure using binary labels $s,t\in\{0,1\}$: $s=0$ is NS and $s=1$ is R around $A$, while $t$ specifies the fermion-parity insertion in the trace. These correspond to theta characteristics $[s/2,t/2]$ on page 14. The usual chiral GSO coefficient is $(-1)^{s+t+st}$.

First take $b=0$. A map wrapping the temporal worldsheet cycle $a$ times transports a spacetime state around the thermal circle $a$ times. A mixed NS–R or R–NS state must therefore acquire $(-1)^a$, while NS–NS and R–R states acquire $+1$. The required factor is $(-1)^{a(s+\bar s)}$. Interchanging the worldsheet cycles exchanges the spin labels as well; modular covariance then requires the partner factor involving $b$. The complete thermal multiplier is

$$(-1)^{a(s+\bar s)+b(t+\bar t)}.$$

In particular, the thermal sum is not obtained by retaining the zero-temperature GSO sum and modifying only the bosonic lattice. After summing the chiral spin structures at zero chemical potential, define

$$K_{ab}(\tau)=\frac{\theta_3^4-(-1)^b\theta_4^4-(-1)^a\theta_2^4}{2\eta^4},\qquad \Xi_{ab}(\tau)=\frac{K_{ab}(\tau)}{\eta^8}.$$

The factor $\eta^{-8}$ accounts for the transverse bosonic oscillators. The odd-spin contribution vanishes in this vacuum trace because of its fermion zero modes. Jacobi’s identity gives the useful nonzero blocks

$(a,b)$ modulo two	$K_{ab}$	Role
$(0,0)$	$0$	The supersymmetric vacuum cancellation
$(1,0)$	$\theta_2^4/\eta^4$	The ungraded one-string Bose-plus-Fermi trace
$(0,1)$	$\theta_4^4/\eta^4$	Odd spatial winding with a reversed NS projection
$(1,1)$	$\theta_3^4/\eta^4$	The other temporal parity in odd spatial winding

This unrefined vacuum expression is the same for IIA and IIB. It does not determine the distinction between their Ramond chiralities: the zero-mode-sensitive spin trace and the chosen GSO projection still carry that information, as explained on page 12.

There is a direct modular check. The transformations of the map labels that preserve their classical action are

$$S:\ (\tau,a,b)\longmapsto(-1/\tau,-b,a),\qquad T:\ (\tau,a,b)\longmapsto(\tau+1,a+b,b).$$

Under $S$, $K$ is taken to the relabelled block without a residual weight. Under $T$, it acquires the common phase $(-1)^{b+1}e^{-i\pi/3}$. The phase cancels against its antiholomorphic conjugate. Meanwhile $\Xi$ has holomorphic modular weight $-4$, so $\tau_2^{-4}|\Xi|^2$ is invariant. Checking the three nonzero blocks is essential; the identity $K_{00}=0$ alone tests none of these factors.

Matsubara momentum and the odd-winding ground state

The torus map action is

$$S_{ab}=\frac{\beta^2}{4\pi\alpha'\tau_2}|a-b\tau|^2.$$

At fixed spin labels, Poisson resummation over $a$ acts on a Gaussian times $(-1)^{a(s+\bar s)}$. If $A=\beta^2/(4\pi^2\alpha'\tau_2)$ and $c=b\tau_1$, the needed identity is

$$\sum_{a\in\mathbb Z}e^{-\pi A(a-c)^2+2\pi iha} =\frac1{\sqrt A}\sum_{n\in\mathbb Z}e^{-\pi(n-h)^2/A-2\pi i(n-h)c},\qquad h=\frac{s+\bar s}{2}.$$

Reversing the summation label puts the momentum in the form $(n+h)/R_0$. Therefore spacetime bosons have integer momentum and spacetime fermions half-integer momentum. This is a consequence of the full spin-dependent sum, rather than an extra imposed rule.

For $b$ odd, the factor $(-1)^{b(t+\bar t)}$ reverses each NS projector. The NS–NS oscillator ground state, projected out at zero temperature, now survives. It has integer thermal momentum and admits $n=0$, with winding $w=b=\pm1$. Its left and right oscillator numbers vanish and level matching holds. Using the sector oscillator intercept $1/2$ on each NS side gives

$$M_{\mathrm{th}}^2=\frac{R_0^2}{\alpha'^2}-\frac2{\alpha'} =\frac{\beta^2-\beta_H^2}{4\pi^2\alpha'^2},\qquad \beta_H=2\pi\sqrt{2\alpha'}.$$

Here $M_{\mathrm{th}}$ is a mass in the nine-dimensional Euclidean spatial problem. It is not the mass of an additional asymptotic Lorentzian particle. The $w=+1$ field and its $w=-1$ conjugate form one complex thermal scalar.

For comparison, the closed bosonic ground state has intercept one on each side. Its analogous winding mass is $R_0^2/\alpha'^2-4/\alpha'$, yielding $\beta_H^{\mathrm{bos}}=4\pi\sqrt{\alpha'}$. The bosonic theory already has a zero-winding tachyon, so this does not define a stable thermal background without an additional prescription.

For flat type II strings the curve is exactly $\alpha'M_{\mathrm{th}}^2=2(x^2-1)$ with $x=\beta/\beta_H$. The two winding orientations form one complex field. Negative mass squared makes its Gaussian saddle unstable. This plot contains no assumption about a stabilizing quartic interaction or the resulting high-temperature phase.

Modular unfolding and the thermal gas

Before unfolding the physical string integrand, it is useful to see the operation in a genuinely convergent example. Use the dimensionless radius $r=R/\sqrt{\alpha'}$ and the invariant circle kernel from page 15:

$$\Gamma_{1,1}(r;\tau)=r\sum_{a,b\in\mathbb Z}e^{-\pi r^2|a-b\tau|^2/\tau_2}.$$

The integral $I(r)=\int_{\mathcal F}d^2\tau\,\Gamma_{1,1}/\tau_2^2$ converges. In the Hamiltonian representation its cusp grows only as $\sqrt{\tau_2}$, so the integrand falls as $\tau_2^{-3/2}$.

The zero map contributes $r\,\operatorname{area}(\mathcal F)=\pi r/3$. Every nonzero pair is $k(c,d)$ with $k>0$ and $(c,d)$ primitive. Modular transformations take a primitive pair to $(1,0)$. Its stabilizer is generated by the translation $\tau\mapsto\tau+1$. Tiling by the corresponding modular images replaces $\mathcal F$ by the strip $\mathcal S=\{-1/2\leq\tau_1<1/2,\ \tau_2>0\}$.

In $\mathrm{PSL}(2,\mathbb Z)$, the pairs $(c,d)$ and $(-c,-d)$ describe the same modular image, but the original lattice sum contains both. Keeping that factor of two, or equivalently summing signed $k$, gives

$$I_{\ne0}(r)=r\sum_{k\ne0}\int_0^\infty\frac{dy}{y^2}e^{-\pi r^2k^2/y} =\frac1{\pi r}\sum_{k\ne0}\frac1{k^2}=\frac{\pi}{3r}.$$

Thus

$$I(r)=\frac\pi3\left(r+\frac1r\right).$$

This nonzero result checks both the modular orbit counting and T-duality. Angelantonj, Florakis, and Pioline, section 3.2 derive the same integral by a different method and explain the need for care with divergent modular integrals. An oscillator-free kernel is a mathematical example; it is not by itself the partition function of a critical two-dimensional string theory.

The normalized type II torus expression

For the flat thermal background, the connected one-loop result is

$$W_1=\frac{\beta V_9}{2(4\pi^2\alpha')^5} \int_{\mathcal F}\frac{d^2\tau}{\tau_2^6} \sum_{a,b\in\mathbb Z}e^{-S_{ab}}|\Xi_{ab}(\tau)|^2.$$

There is one ten-dimensional constant-mode normalization, with Euclidean volume $\beta V_9$. The factor $1/2$ is the usual real-field one-loop normalization; its consistency will be checked by matching the gas. There is no second independent momentum Gaussian to multiply into this expression. The spin-dependent map sum already represents the thermal coordinate. Its zero orbit vanishes here by $\Xi_{00}=0$.

For $\beta>\beta_H$ in the noncompact flat background, the thermal integral and its orbit rearrangement converge. Applying the preceding unfolding argument, including the two orientations, gives

$$W_1=\frac{\beta V_9}{(4\pi^2\alpha')^5} \sum_{r=1}^{\infty}\int_{\mathcal S}\frac{d^2\tau}{\tau_2^6} e^{-r^2\beta^2/(4\pi\alpha'\tau_2)}|\Xi_{r0}(\tau)|^2.$$

The notation $r$ in this formula is the positive wrapping number, rather than the dimensionless radius used in the example above. At even $r$ the spin sum vanishes. At odd $r$, $\Xi_{r0}=\theta_2^4/\eta^{12}$, whose leading absolute square is $256$: the $128$ bosonic and $128$ fermionic massless physical polarizations.

The $\tau_1$ integral projects onto level-matched states. At every mass level, the oscillator trace weights the states by $d_{\mathrm B}-(-1)^r d_{\mathrm F}$. Setting $s=\pi\alpha'\tau_2$ consequently gives

$$\frac{W_1}{\beta V_9}=\frac1{(4\pi)^5}\sum_{r=1}^{\infty}\sum_{\text{mass levels}} \bigl[d_{\mathrm B}+(-1)^{r+1}d_{\mathrm F}\bigr] \int_0^\infty\frac{ds}{s^6}e^{-sM^2-r^2\beta^2/(4s)}.$$

This is exactly $-F_1/V_9$ from the particle gas. In a supersymmetric zero-temperature spectrum $d_{\mathrm B}=d_{\mathrm F}$, explaining why only odd wrapping survives. Thermal supersymmetry is nevertheless broken: the result adds bosons and fermions at odd wrapping, rather than cancelling them.

Unfolding is not a prescription for assigning a value to a divergent tachyonic integral. One must first specify a common convergent domain or a regulator and its continuation. A cusp cutoff in $\mathcal F$ maps to a collection of regions in the strip, not to the simple horizontal cutoff $\tau_2<\Lambda$. Exercise 2 makes this distinction explicit.

Exponential degeneracy and the Hagedorn scale

At large oscillator level $N$, the transverse chiral theory has effective central charge $c=12$ for type II strings and $c=24$ for bosonic strings. The leading Cardy growth is

$$\log d_N\sim2\pi\sqrt{\frac{cN}{6}}.$$

For a level-matched closed string, multiply the two chiral degeneracies and use $M\sim2\sqrt{N/\alpha'}$. This yields

$$\log d_{\mathrm{closed}}(M)\sim\beta_H M,\qquad \beta_H=2\pi\sqrt{\frac{c\alpha'}6}.$$

The two values of $c$ reproduce the winding thresholds derived above. The mass-level degeneracy is not yet a density per unit energy: changing variables from $N$ to $M$ and integrating center-of-mass momentum supply additional powers. Write the resulting large-energy one-string density schematically as

$$\rho_1(E)\sim C E^{-a}e^{\beta_H E}.$$

Then $z_1(\beta)$ contains $\int^\infty dE\,E^{-a}e^{-(\beta-\beta_H)E}$. For $\beta>\beta_H$ the tail converges; for $\beta<\beta_H$ it diverges exponentially. At equality, the power matters: the trace converges if $a>1$, and its $n$th energy moment converges only if $a>n+1$. A finite value at the threshold need not be analytic there.

The plotted model is $f(x)=\log[\rho_1(E)e^{-\beta E}/(\rho_1(E_0)e^{-\beta E_0})]=-(11/2)\log x-\delta(x-1)$, with $x=E/E_0$ and $\delta=(\beta-\beta_H)E_0$. The power $11/2$ is appropriate to the flat nine-dimensional asymptotic tail derived below, but the finite-range curves are illustrative. At $\delta=-0.3$ the minimum is $x=55/3$; eventual growth does not require the curve to have become positive within the plotted interval.

The equality between an enormous oscillator sum and a light Euclidean winding state is a channel relation. It does not mean that the massive strings have disappeared from the spectrum. Their collective contribution has a simpler description in a different degeneration of the worldsheet.

The thermal scalar and infrared singularities

Near $\beta_H$ from above, retain $b=\pm1$ and the NS–NS ground state in the fundamental-domain cusp. Poisson resummation of the temporal winding gives the zero-momentum prefactor

$$\frac{2\pi\sqrt{\alpha'\tau_2}}{\beta}.$$

The ground-state exponential is $e^{-\pi\alpha'\tau_2M_{\mathrm{th}}^2}$. The two winding orientations cancel the torus factor $1/2$. With $s=\pi\alpha'\tau_2$, the singular part therefore becomes

$$\frac{W_{\mathrm{sing}}}{V_d}=\frac1{(4\pi)^{d/2}} \int_{s_0}^{\infty}\frac{ds}{s^{1+d/2}}e^{-m^2s},\qquad m^2=M_{\mathrm{th}}^2>0,$$

where $d=9$ in the flat type II example. The cutoff $s_0>0$ restricts the approximation to long proper time. Shorter tubes and heavier states contribute terms analytic near $m^2=0$. This is the heat-kernel representation of $W_{\mathrm{sing}}=-\operatorname{Tr}\log(-\nabla^2+m^2)$ for one complex scalar. Inserting another $1/2$ would incorrectly count only one winding orientation.

For $d>0$, the integral itself is finite at $m^2=0$. Its $n$th derivative with respect to $m^2$ contains $s^{n-1-d/2}$ and diverges at the upper endpoint when $n\geq d/2$, logarithmically at equality. Up to analytic terms, for non-even $d$,

$$\frac{F_{\mathrm{sing}}}{V_d}=-\frac{\Gamma(-d/2)}{\beta(4\pi)^{d/2}}(m^2)^{d/2}.$$

Even $d$ gives the corresponding $(m^2)^{d/2}\log m^2$ term. In $d=9$, the first nonanalytic term has power $9/2$ and the fifth derivative diverges. The temperature dependence of $m^2$ has a nonzero first derivative at $\beta_H$, so the same leading threshold order appears in derivatives with respect to $\beta$.

The density prefactor also follows without guessing an oscillator power. Let $\delta\beta=\beta-\beta_H$ and $A=\beta_H/(2\pi^2\alpha'^2)$, so that $m^2=A\delta\beta+O(\delta\beta^2)$. Substituting $E=As$ in the long-time integral gives

$$\rho_1(E)\sim V_d\left(\frac{A}{4\pi}\right)^{d/2}\frac{e^{\beta_HE}}{E^{1+d/2}}.$$

The singular part of $W$ equals that of the $r=1$ trace, since $r\geq2$ evaluates the high-energy density at a strictly larger inverse temperature. Thus $a=1+d/2$, giving $11/2$ for $d=9$. This argument, including its dependence on spatial size, is discussed in Horowitz and Polchinski, section 2.

Finite volume and the unstable Gaussian

In a periodic spatial box, replace $V_d(4\pi s)^{-d/2}$ by $\sum_{\mathbf k}e^{-s\mathbf k^2}$. At sufficiently large $s$, the isolated zero mode dominates:

$$W_{\mathrm{sing}}\sim\int_{s_0}^{\infty}\frac{ds}{s}e^{-m^2s}\sim-\log(m^2s_0),\qquad F_{\mathrm{sing}}\sim\frac1\beta\log(m^2s_0).$$

The threshold behavior is therefore different if the finite-volume limit is kept until the end. This statement concerns the additional thermal-scalar singularity; an exactly zero-energy massless boson in a free finite box already requires a prescription for its occupation zero mode. One can fix those background collective modes or regulate them separately without changing the winding-state analysis.

If the thermal-scalar operator develops negative spectrum, the Gaussian integral about $\chi=0$ is unstable. Analytically continuing its determinant can diagnose that instability; it does not construct a new equilibrium state. Interactions, massless-field exchange, and gravitational backreaction can change the phase structure or cause a transition before the free threshold. A positive quartic potential and a continuous transition do not follow from the mass formula alone.

In an AdS/CFT setting, deconfinement also involves comparing different bulk saddles with the same boundary data. At fixed ‘t Hooft coupling, the connected genus-one gas contribution is of order $N^0$, whereas a classical black-hole contribution can be of order $N^2$, since its action scales as $g_s^{-2}$. The black D3-brane thermodynamics on page 28 develops the latter scaling. A full Hawking–Page comparison requires the competing free energies and the chosen boundary geometry; the winding mass around one background is not that comparison.

Thermal winding in a static curved background

Consider a slowly varying, static string-frame background

$$ds_S^2=G_{00}(x)dY^2+g_{ij}(x)dx^i dx^j,\qquad Y\sim Y+\beta,$$

with $G_{00}>0$, no mixed time-space components, and vanishing Kalb–Ramond field. The dilaton $\Phi$ is the field in the global frame definitions. The locally measured circumference is $\beta_{\mathrm{loc}}=\beta\sqrt{G_{00}}$, but the mode also pays a spatial gradient energy.

At leading order in the derivative expansion, the winding action has the form

$$S_{\mathrm{th}}=\beta\int d^d x\,\mu(x) \left[g^{ij}\partial_i\chi^*\partial_j\chi+M_{\mathrm{loc}}^2(x)|\chi|^2\right],$$ $$\mu=\sqrt g\,\sqrt{G_{00}}e^{-2\Phi},\qquad M_{\mathrm{loc}}^2=\frac{\beta^2G_{00}-\beta_H^2}{4\pi^2\alpha'^2}.$$

An overall positive field normalization has been absorbed into $\chi$. The factor $\sqrt{G_{00}}$ belongs to the full spacetime volume element before reducing on the thermal circle. It cannot be removed while leaving the kinetic operator unchanged. One way to derive the winding term is to reduce a momentum-mode action and apply the local T-duality rules, temporarily suppressing a radius-dependent constant in the dilaton shift: $G_{00}\mapsto G_{00}^{-1}$, $\Phi\mapsto\Phi-\tfrac12\log G_{00}$. The measure $\sqrt{G_{00}}e^{-2\Phi}$ is invariant under this local part of the transformation. The full shift also includes $-\log(R_0/\sqrt{\alpha'})$, as on page 15; its constant factor can be absorbed into the field normalization here. The momentum contribution becomes $w^2R_0^2G_{00}/\alpha'^2$. Selecting $w=\pm1$ gives the displayed action. See Horowitz and Polchinski, equation (3.1) and Mertens, Verschelde, and Zakharov, section 5.

Integration by parts gives the self-adjoint operator in the measure $\mu\,d^d x$:

$$\mathcal H_{\mathrm{th}}=-\frac1\mu\partial_i\bigl(\mu g^{ij}\partial_j\bigr)+M_{\mathrm{loc}}^2.$$

For boundary conditions that remove surface terms, the bottom of its spectrum satisfies

$$\lambda_0=\inf_{\chi\ne0} \frac{\int d^d x\,\mu\,[g^{ij}\partial_i\chi^*\partial_j\chi+M_{\mathrm{loc}}^2|\chi|^2]} {\int d^d x\,\mu\,|\chi|^2}.$$

A negative value of $M_{\mathrm{loc}}^2$ somewhere is not sufficient to make this quotient negative. Conversely, a trial function with a negative quotient proves a negative spectral contribution and an unstable quadratic direction. Calling it a negative eigenvalue additionally requires a discrete mode or suitable bound-state assumptions; on noncompact space the bottom may belong to a continuum. Exercise 6 gives a confining example with a discrete spectrum.

These statements describe a leading quadratic operator in a specified background. Curvature and higher-derivative corrections, extra light modes, boundary conditions, and interactions must be controlled before identifying an exact string threshold. In particular, a thermal circle that contracts at a smooth Euclidean horizon is not globally the noncontractible circle used in the flat winding argument. The Tolman redshift by itself does not prove a winding tachyon at every horizon or determine the black-hole phase diagram.

Exercises

Exercise 1: Occupations, an index, and a normalized gas

For a bosonic and a fermionic oscillator of the same energy $E>0$, compute their ordinary and graded partition functions, derive the connected occupation expansion, and explain the Euclidean fermion boundary sign. Then evaluate the massless free energy per polarization in three spatial dimensions from the proper-time formula. Keep the Bose and Fermi numerical coefficients.

Solution

Writing $q=e^{-\beta E}$, the separate ordinary traces are $Z_{\mathrm B}=(1-q)^{-1}$ and $Z_{\mathrm F}=1+q$. Their product is $(1+q)/(1-q)$, so

$$\log Z_{\mathrm{pair}}=2\sum_{\substack{r\geq1\\r\ \mathrm{odd}}}\frac{q^r}{r}.$$

The fermionic graded trace is $1-q$. The paired graded trace is therefore one: the zero-energy vacuum is unpaired, while the positive-energy states cancel. An offset in the vacuum energy or a nonconvergent continuum requires separate analysis; this finite example is not a general proof of index independence.

For a fermionic coherent state $|\eta\rangle$, the ordinary trace may be written as a Grassmann integral of $\langle-\eta|e^{-\beta H}|\eta\rangle$. Its endpoint minus sign gives antiperiodic fields. Inserting $(-1)^{\mathsf F}$ changes the endpoint to $\langle\eta|$, giving periodic fields. Fourier expansion then produces half-integer or integer Matsubara frequencies, respectively.

For $d=3$ and $M=0$, put $u=r^2\beta^2/(4s)$ in the proper-time integral:

$$\int_0^\infty\frac{ds}{s^3}e^{-r^2\beta^2/(4s)}=\frac{16}{r^4\beta^4}\int_0^\infty u e^{-u}du=\frac{16}{r^4\beta^4}.$$

Multiplication by $-(4\pi)^{-2}$ yields

$$\frac{F_{\mathrm B}}{V_3}=-\frac{\zeta(4)}{\pi^2\beta^4}=-\frac{\pi^2}{90\beta^4},\qquad \frac{F_{\mathrm F}}{V_3}=-\frac{(1-2^{-3})\zeta(4)}{\pi^2\beta^4}=-\frac{7\pi^2}{720\beta^4}.$$

These are per physical polarization. The alternating sum gives the familiar $7/8$ ratio. A Dirac field requires its full particle, antiparticle, and spin multiplicity; it is not one polarization.

Exercise 2: A cutoff does not commute with naive unfolding

For the convergent circle kernel, redo the nonzero-orbit integral with only the unfolded strip restricted to $0<y<L$. Determine its exact sum and its leading difference from the unrestricted answer. Explain why this difference is not the tail removed by cutting the original fundamental domain at $\tau_2=L$.

Solution

With $A_k=\pi r^2k^2$, substitution $u=A_k/y$ gives

$$\int_0^L\frac{dy}{y^2}e^{-A_k/y}=\frac{e^{-A_k/L}}{A_k}.$$

The truncated nonzero strip sum is consequently

$$I_{\ne0}^{\mathcal S}(L)=\frac1{\pi r}\sum_{k\ne0}\frac{e^{-\pi r^2k^2/L}}{k^2}.$$

Let $a=\pi r^2/L$ and $D(a)=\sum_{k\ne0}(1-e^{-ak^2})/k^2$. Differentiating before taking $a\to0$ gives

$$D'(a)=\sum_{k\ne0}e^{-ak^2}=\sqrt{\frac\pi a}-1+O(a^{-1/2}e^{-\pi^2/a}).$$

Using $D(0)=0$ and integrating,

$$I_{\ne0}(r)-I_{\ne0}^{\mathcal S}(L)=\frac2{\sqrt L}-\frac rL+O(e^{-\pi L/r^2}).$$

By contrast, the complete Hamiltonian kernel has cusp behavior $\Gamma_{1,1}\sim\sqrt y$ at fixed $r$, so its fundamental-domain tail is $2/\sqrt L$ plus exponentially small terms. The zero orbit alone has tail $r/L$. These different allocations are expected: modular images of a horizontal cutoff are curved regions, and separating the orbits before imposing a new rectangular cutoff changes which points are omitted.

Both unregulated expressions converge to $\pi(r+r^{-1})/3$, and the two opposite primitive orientations are indispensable for that result. For a tachyonic oscillator kernel, neither taking the cutoff to infinity nor exchanging the sum and integral is justified in the same way. One must specify a common regulator or continuation rather than equate independently truncated divergent integrals.

Exercise 3: Resolve the thermal projection before the vacuum trace vanishes

Work at fixed left and right spin labels. Show how temporal-winding parity gives Bose or Fermi thermal momentum, and how odd spatial winding changes the NS projector. Use the first NS excitations to identify which low states survive. Finally verify the $T$ transformation of the nonzero chiral blocks, including its common phase.

Solution

Set $h=(s+\bar s)/2$ in the Poisson identity in the main text. The resulting momentum is $p_0=(n+h)/R_0$. If $(s,\bar s)=(0,0)$ or $(1,1)$, the shift is integral and can be absorbed into $n$. If exactly one sector is R, the shift is half-integral. This proves the spin-statistics assignment in the thermal spectrum directly from the phase.

In a chiral NS sector the usual GSO sum is proportional to $\theta_3^4-\theta_4^4$. In an occupation basis with $N_\psi$ fermion creation operators, it retains odd $N_\psi$. The extra $(-1)^{bt}$ changes the projector to

$$P_{\mathrm{NS}}^{(b)}=\frac12\left[1-(-1)^b(-1)^{N_\psi}\right].$$

For even $b$, the vacuum is removed and the eight $\psi^i_{-1/2}|0\rangle$ states survive. For odd $b$, the vacuum survives and those eight states are removed. At chiral level one, the eight $\alpha^i_{-1}|0\rangle$ states and the $\binom82=28$ states $\psi^i_{-1/2}\psi^j_{-1/2}|0\rangle$ survive the odd-winding projector. Pairing two NS vacua gives the thermal ground state at $n=0$, $b=\pm1$. Pairing higher states additionally requires the usual momentum-winding level matching.

The theta transformations are $\theta_3\mapsto\theta_4$, $\theta_4\mapsto\theta_3$, $\theta_2\mapsto e^{i\pi/4}\theta_2$, and $\eta\mapsto e^{i\pi/12}\eta$ under $T$. Thus

$$K_{ab}(\tau+1)=(-1)^{b+1}e^{-i\pi/3}K_{a+b,b}(\tau).$$

The integer action uses the simultaneous map $a\mapsto a+b$, as in the main text; because $K$ depends only on parity, either direction of this relabelling gives the displayed identity. The common phase drops out of $|K|^2$. Under $S$, the exchange of $\theta_2$ and $\theta_4$ exchanges the two winding parities. The surviving zero-momentum odd-winding state is therefore compatible with modular covariance, not an isolated ad hoc reversal of one projector.

The unrefined $\theta_1(0)=0$ trace cannot distinguish the two Ramond chiralities. Keeping that information requires the Ramond zero-mode projection or a refined trace; it cannot be reconstructed by assigning a nonzero value to a vanishing vacuum character.

Exercise 4: Obtain the power prefactor from oscillator asymptotics

Use the transverse chiral type II generating function $f(\tau)=\theta_2(\tau)^4/\eta(\tau)^{12}$. Determine its small-$t$ behavior at $\tau=it$, estimate its coefficient growth by a saddle point, and recover the power $E^{-11/2}$ after level matching and center-of-mass integration. Explain why the full gas has the same leading Hagedorn singularity as its first occupation term.

Solution

Modular transformation gives

$$f(it)=t^4\frac{\theta_4(i/t)^4}{\eta(i/t)^{12}}\sim t^4e^{\pi/t},\qquad t\to0^+.$$

For $f(it)=\sum_N d_Ne^{-2\pi tN}$, the inverse-coefficient integral has exponent $2\pi Nt+\pi/t$. Its stationary point is $t_*=(2N)^{-1/2}$, and its exponential is $e^{2\pi\sqrt{2N}}$. The factor $t_*^4$ gives $N^{-2}$, while the Gaussian width of the steepest-descent contour gives $N^{-3/4}$. Hence, up to a constant,

$$d_N\sim N^{-11/4}e^{2\pi\sqrt{2N}}.$$

At a level-matched closed level, square this result. Since $N\sim\alpha'M^2/4$ and $dN/dM\propto M$, the density per unit rest mass is $M^{-10}e^{\beta_HM}$.

For a very massive string with momenta relevant to the thermal trace, $\sqrt{M^2+\mathbf p^2}=M+\mathbf p^2/(2M)+\cdots$. The Gaussian momentum integral in nine spatial dimensions is

$$\int\frac{d^9p}{(2\pi)^9}e^{-\beta\mathbf p^2/(2M)}=\left(\frac{M}{2\pi\beta}\right)^{9/2}.$$

It changes the power in the high-mass trace to $M^{-11/2}$. Equivalently, at fixed large total energy $E$, write $M=E-\mathbf p^2/(2E)+\cdots$ inside $e^{\beta_H M}$. The resulting momentum Gaussian again contributes $E^{9/2}$, and the Jacobian $dM/dE$ tends to one. This gives $\rho_1(E)\propto E^{-11/2}e^{\beta_HE}$, in agreement with the thermal-scalar inverse Laplace transform.

At $\beta=\beta_H$, the one-string trace and its first four energy moments have convergent tails. The fifth does not. For every fixed $r\geq2$, the occupation term uses $r\beta$, so its exponential suppression remains strict in a neighborhood of $\beta_H$. The leading nonanalyticity of $W_{\mathrm{gas}}$ therefore comes from $r=1$. The remaining terms are still necessary for the complete free energy and its low-energy Bose/Fermi statistics.

Exercise 5: A box changes the critical singularity

Put the complex thermal scalar in a periodic cube of side $L$ and let $\varepsilon=m^2>0$. Derive both the momentum and image expressions for its heat trace. Identify the crossover between the finite-box logarithm and the noncompact power law. Compute the leading singular fifth derivative of $F/V_9$ in nine noncompact dimensions, and recover the finite-box high-energy density.

Solution

For a periodic cube,

$$K_L(s)=\sum_{\mathbf n\in\mathbb Z^d}e^{-4\pi^2s|\mathbf n|^2/L^2} =\frac{L^d}{(4\pi s)^{d/2}}\sum_{\mathbf w\in\mathbb Z^d}e^{-L^2|\mathbf w|^2/(4s)}.$$

The equality follows by Poisson resummation in every spatial direction. If $s\ll L^2$, the zero image dominates and the heat trace is approximately the noncompact volume factor. If $s\gg L^2$, the zero momentum mode dominates and $K_L(s)\to1$.

The singular determinant is $W=\int_{s_0}^{\infty}ds\,e^{-\varepsilon s}K_L(s)/s$. Its relevant large-time scale is $s\sim\varepsilon^{-1}$. Thus $\varepsilon L^2\gg1$ gives the noncompact regime, whereas $\varepsilon L^2\ll1$ gives the finite-box logarithm. For the latter,

$$W\sim E_1(\varepsilon s_0)=-\gamma_E-\log(\varepsilon s_0)+O(\varepsilon s_0).$$

This is one complex mode, so the logarithm has coefficient one. The other eigenvalues remain separated by a gap of order $L^{-2}$.

In the noncompact $d=9$ case, differentiate the regulated proper-time integral five times with respect to $\varepsilon$, keeping the smooth factor $1/\beta$ fixed when quoting the leading singular term:

$$\frac{\partial^5(F/V_9)}{\partial\varepsilon^5} \sim\frac1{\beta(4\pi)^{9/2}}\int_0^\infty ds\,s^{-1/2}e^{-\varepsilon s} =\frac{\sqrt\pi}{\beta(4\pi)^{9/2}}\varepsilon^{-1/2}.$$

Changing the lower limit to zero here affects only terms that are less singular. Lower derivatives of the nonanalytic term vanish at the threshold, while analytic contributions from shorter times remain.

Finally, $\varepsilon\simeq A(\beta-\beta_H)$ converts the finite-box logarithm into $\int^\infty dE\,e^{-(\beta-\beta_H)E}/E$. Thus the finite-box asymptotic density has the power $\rho_1(E)\sim e^{\beta_HE}/E$, instead of the noncompact $E^{-1-d/2}$. Sending $L\to\infty$ before the threshold and keeping $L$ finite until afterward therefore give different singular asymptotics. Neither manipulation determines the interacting phase for $\varepsilon<0$.

Exercise 6: The measure can change the lowest eigenvalue

Write the spatial measure as $\mu=\sqrt g\,e^{2h}$ and derive a unitarily equivalent operator in the measure $\sqrt g\,d^dx$. Then consider the one-dimensional model $g=1$, $h(x)=a x^2/2$, and $M_{\mathrm{loc}}^2=m_0^2+\kappa^2x^2$, where $a$ and $\kappa>0$ have dimensions of inverse length squared. Find the exact lowest eigenvalue and explain why the pointwise sign of $m_0^2$ is insufficient. This is an operator model, not an assertion that these functions solve string background equations.

Solution

The original kinetic operator is $-\Delta_g-2\nabla h\cdot\nabla$. Set $\chi=e^{-h}\psi$. This map is unitary because $\int\mu|\chi|^2=\int\sqrt g|\psi|^2$. Direct differentiation gives

$$e^h\mathcal H_{\mathrm{th}}e^{-h}=-\Delta_g+M_{\mathrm{loc}}^2+(\nabla h)^2+\Delta_g h.$$

The last two terms are required by the measure; omitting them changes the spectrum. In the example they give

$$\widetilde{\mathcal H}=-\frac{d^2}{dx^2}+(\kappa^2+a^2)x^2+m_0^2+a.$$

Let $\omega=\sqrt{\kappa^2+a^2}$. The normalized-measure eigenfunctions are harmonic-oscillator wavefunctions, with eigenvalues

$$\lambda_n=(2n+1)\omega+m_0^2+a,\qquad \chi_0(x)\propto e^{-(a+\omega)x^2/2}.$$

Even if $\chi_0$ grows in an alternative example with a different measure, it is its weighted norm that would decide admissibility. Here $\kappa>0$ ensures $\omega>|a|$, so $\chi_0$ itself decays and is normalizable in the weighted norm.

The instability criterion is $m_0^2<-a-\sqrt{\kappa^2+a^2}$. For instance, at $a=0$ and $m_0^2=-\kappa/2$, the local mass is negative near the origin but $\lambda_0=\kappa/2>0$. Localization costs gradient energy. Conversely, the exact Gaussian trial function gives a negative Rayleigh quotient once the stated threshold is crossed.

For the actual leading thermal action, $h=\tfrac14\log G_{00}-\Phi$. Curvature and dilaton gradients therefore affect the operator in addition to the local circumference. Applying this model to a string background would require checking its equations, derivative expansion, boundary conditions, and global winding sector first.

References

• Angelantonj, C.; Florakis, I.; Pioline, B. “A new look at one-loop integrals in string theory.” Communications in Number Theory and Physics 6 (2012), 159–201. doi:10.4310/CNTP.2012.v6.n1.a4, arXiv:1110.5318. Section 3.2 evaluates the convergent one-circle modular integral; the surrounding discussion treats convergence and regularization.
• Horowitz, G. T.; Polchinski, J. “Self Gravitating Fundamental Strings.” Physical Review D 57 (1998), 2557–2563. doi:10.1103/PhysRevD.57.2557, arXiv:hep-th/9707170. Section 2 derives the thermal-scalar density and volume dependence; equation (3.1) gives the static metric and dilaton action. Its subsequent self-interacting single-string analysis is distinct from a complete thermal-gas phase diagram.
• Kruczenski, M.; Lawrence, A. “Random walks and the Hagedorn transition.” Journal of High Energy Physics 07 (2006), 031. doi:10.1088/1126-6708/2006/07/031, arXiv:hep-th/0508148. Sections 2–3 connect the gas, modular channels, long strings, and a thermal winding operator.
• Mertens, T. G.; Verschelde, H.; Zakharov, V. I. “Near-Hagedorn Thermodynamics and Random Walks: a General Formalism in Curved Backgrounds.” Journal of High Energy Physics 02 (2014), 127. doi:10.1007/JHEP02(2014)127, arXiv:1305.7443. Sections 2–3 discuss the thermal worldsheet and flat-space comparison; section 5 derives the leading winding action and identifies measure and higher-derivative qualifications.