The Witten Index, Vacuum Counting, and Its Failure Modes

The Witten index is a signed count of supersymmetric ground states, not an ordinary count and not automatically a well-defined trace. Under precise trace-class or Fredholm hypotheses it is independent of Euclidean time and of supersymmetry-preserving deformations. A nonzero value proves that supersymmetry is unbroken; a zero value proves nothing by itself.

Required background. The spectral-pairing theorem supplies the cancellation of positive-energy states. Helpful background. The Fredholm index gives the operator-theoretic formulation, while Morse deformation and the continuum and boundary analysis exhibit the principal ways a naive counting argument can fail.

The graded heat trace

Let

$$\mathcal H=\mathcal H_{\bar 0}\oplus\mathcal H_{\bar 1}, \qquad \Gamma=(-1)^F, \qquad H=\frac12\{\mathcal Q,\mathcal Q^\dagger\}\geq 0,$$

with $H$ self-adjoint, $\Gamma H=H\Gamma$, and $\Gamma\mathcal Q=-\mathcal Q\Gamma$. For $\beta>0$, define

$$I(\beta) =\operatorname{Str}_{\mathcal H}e^{-\beta H} \equiv \operatorname{Tr}_{\mathcal H} \bigl(\Gamma e^{-\beta H}\bigr) =\operatorname{Tr}_{\mathcal H_{\bar0}}e^{-\beta H_{\bar0}} -\operatorname{Tr}_{\mathcal H_{\bar1}}e^{-\beta H_{\bar1}}.$$

This definition is literal if $e^{-\beta H}$ is trace class. More generally, one may define a relative supertrace after specifying a regulator and proving that the even–odd difference has a regulator-independent limit. Without one of these statements, the displayed expression is only formal.

For the closed differential $A:\mathcal D(A)\subset\mathcal H_{\bar0}\to\mathcal H_{\bar1}$ used throughout this chapter,

$$H_{\bar0}=\frac12A^\dagger A, \qquad H_{\bar1}=\frac12AA^\dagger.$$

If $A$ is Fredholm, its range is closed and its kernel and cokernel are finite dimensional. Its analytic index is

$$\operatorname{ind}A =\dim\ker A-\dim\ker A^\dagger.$$

Whenever the graded heat trace exists and the positive spectrum is paired without a contribution from infinity,

$$\boxed{ I(\beta) =n_{\bar0}^{(0)}-n_{\bar1}^{(0)} =\operatorname{ind}A . }$$

The first equality is the spectral version of the Fredholm index theorem in this setting; the assumptions that make it legitimate are as important as the formula. See Cooper, Khare, and Sukhatme 1995, §2.2, arXiv PDF pp. 23–25 and Weinberg 2000, §29.1, pp. 248–255.

Why positive energies cancel

Suppose first that $H$ has discrete spectrum of finite multiplicity and that the heat trace converges. The pairing maps

$$A:\ker(H_{\bar0}-E)\longrightarrow\ker(H_{\bar1}-E), \qquad A^\dagger:\ker(H_{\bar1}-E)\longrightarrow\ker(H_{\bar0}-E)$$

are inverse up to the factor $2E$ for every $E>0$. Thus $\dim\ker(H_{\bar0}-E)=\dim\ker(H_{\bar1}-E)$, and the contribution of that eigenspace to $I(\beta)$ is zero. Only

$$\ker H_{\bar0}=\ker A, \qquad \ker H_{\bar1}=\ker A^\dagger$$

survives. This eigenbasis proof is often safer than manipulating unbounded supercharges inside a trace.

There is also a useful formal calculation. Where differentiation and graded cyclicity are justified,

$$\begin{aligned} \frac{dI}{d\beta} &=-\operatorname{Str}\!\left(He^{-\beta H}\right)\\ &=-\frac12\operatorname{Str}\!\left( \mathcal Q\mathcal Q^\dagger e^{-\beta H} +\mathcal Q^\dagger\mathcal Qe^{-\beta H}\right)=0. \end{aligned}$$

The last equality uses that $\mathcal Q$ is odd and commutes with $H$: graded cyclicity changes the sign of one term and cancels the other. For unbounded $\mathcal Q$, one must first know that the relevant products admit trace-class closures. The slogan “the supertrace of a supercommutator vanishes” is not a substitute for that analytic check.

Deformation invariance and its hypotheses

Let $\mathcal Q_s$ be a differentiable family of closed odd operators with a fixed grading, and set

$$H_s=\frac12\{\mathcal Q_s,\mathcal Q_s^\dagger\}.$$

On a common invariant core,

$$\partial_sH_s =\frac12\bigl( \{\partial_s\mathcal Q_s,\mathcal Q_s^\dagger\} +\{\mathcal Q_s,\partial_s\mathcal Q_s^\dagger\} \bigr).$$

If the domains vary smoothly in a controlled quadratic-form sense, the heat-kernel derivatives are trace class, and a uniform trace bound permits Duhamel differentiation, then

$$\partial_s I_s(\beta) =-\beta\,\operatorname{Str}\!\left( (\partial_sH_s)e^{-\beta H_s}\right)=0.$$

Equivalently, a gap-continuous family of closed Fredholm differentials—or a norm-continuous family of their bounded transforms—has constant Fredholm index. These are two versions of the same stability statement. The original use of this protected count to constrain supersymmetry breaking is developed in Witten 1982, §§1–2, pp. 253–271.

The conclusion can fail when any hypothesis fails:

• an asymptotic mass tends to zero and $A_s$ ceases to be Fredholm;
• a normalizable zero mode escapes to infinity or merges into a continuum;
• the boundary condition, and hence the operator domain, changes;
• the zero-energy kernel becomes infinite dimensional;
• the regulated even and odd traces exist separately only after incompatible subtractions; or
• a limit in volume, $\beta$, or deformation parameter is interchanged without uniform convergence.

The Gaussian deformation on the line is a concrete warning: conjugation by an unbounded function can create an $L^2$ zero mode even though the deformed complexes are algebraically isomorphic on compactly supported forms.

Three models with two index values

The index distinguishes a signed imbalance, not the total number of vacua.

Model	Even zero modes	Odd zero modes	Index	Supersymmetry
Harmonic superpotential $w(x)=\omega x$, $\omega>0$	1	0	$+1$	Unbroken
de Rham complex of $S^1$	1	1	$0$	Unbroken
$w(x)=x^2-a^2$ on $\mathbb R$	0	0	$0$	Broken

For $S^1$, the constant function and constant one-form are both harmonic. Their contributions cancel, so the index is the Euler characteristic $\chi(S^1)=0$ even though there are two supersymmetric vacua.

For $w=x^2-a^2$, write

$$W(x)=\frac{x^3}{3}-a^2x, \qquad A=\frac{d}{dx}+W'(x).$$

The formal even and odd zero modes are $e^{-W}$ and $e^{W}$. Each decays at one end of the line and grows at the other, so neither lies in $L^2(\mathbb R)$. The zero index now accompanies broken supersymmetry. Thus the same value $I=0$ realizes opposite physics.

What vacuum counting can establish

When the index is defined,

$$n_{\mathrm{vac}} =n_{\bar0}^{(0)}+n_{\bar1}^{(0)} \geq \left\lvert n_{\bar0}^{(0)}-n_{\bar1}^{(0)}\right\rvert =\lvert I\rvert.$$

Consequently:

• $I\neq0$ guarantees at least $\lvert I\rvert$ supersymmetric vacua and rules out spontaneous supersymmetry breaking.
• $I=0$ allows either no supersymmetric vacuum or equal nonzero numbers of even and odd vacua.
• an undefined or regulator-dependent “index” supports neither conclusion.

The index can remain constant while the total number of ground states changes: an even–odd pair may meet zero energy or leave it together. Detecting that extra information requires the cohomology groups themselves, a refined index, or additional quantum numbers—not the ordinary signed trace.

Periodic fermions, not a thermal partition function

In a Euclidean path-integral representation, taking the trace identifies the bosonic endpoint. Inserting $\Gamma=(-1)^F$ changes the fermionic gluing from antiperiodic to periodic:

$$I(\beta) =\int_{\substack{x(\beta)=x(0)\\ \psi(\beta)=\psi(0)}}\mathcal D x\,\mathcal D\psi\; e^{-S_E[x,\psi]}.$$

Periodic fermions preserve the constant supersymmetry parameter and expose fermion zero modes. This is why localization and semiclassical deformation can compute the index. It is also why $I(\beta)$ should not be called a thermal partition function: the physical thermal trace has antiperiodic fermions. The boundary-condition distinction is derived in Cooper, Khare, and Sukhatme 1995, §8, arXiv PDF pp. 79–85.

Continuous spectra require a relative statement

If scattering states are present, the even and odd heat traces are generally infinite. A regulator may leave

$$I_{\mathrm{reg}}(\beta) =n_{\bar0}^{(0)}-n_{\bar1}^{(0)} +\int_{E_{\mathrm{th}}}^{\infty} e^{-\beta E} \bigl[\rho_{\bar0}(E)-\rho_{\bar1}(E)\bigr]\,dE.$$

The density difference is a boundary-at-infinity effect invisible to the normalizable eigenstate pairing argument. It can make the regulated answer $\beta$ dependent or contribute a threshold term. Its value depends on the specified comparison operator, scattering normalization, and order of limits. The next page derives the phase-shift version and separates this continuum effect from changes caused by a physical boundary.

Check your understanding

Consider a finite-dimensional graded space with two even states and three odd states. Suppose the supercharge has

$$A= \begin{pmatrix} 1&0 \\ 0&0 \\ 0&0 \end{pmatrix} : \mathbb C^2_{\bar0}\longrightarrow\mathbb C^3_{\bar1}.$$

Find the ground-state degeneracies, the Witten index, and the positive-energy spectrum of $H=\frac12\operatorname{diag}(A^\dagger A,AA^\dagger)$.

Solution

The kernel of $A$ is spanned by the second even basis vector, so $n_{\bar0}^{(0)}=1$. The kernel of $A^\dagger$ is spanned by the second and third odd basis vectors, so $n_{\bar1}^{(0)}=2$. Hence

$$I=1-2=-1.$$

Both $A^\dagger A$ and $AA^\dagger$ have one nonzero eigenvalue equal to one. Thus $H$ has a paired even–odd level at $E=1/2$. Its contributions $e^{-\beta/2}-e^{-\beta/2}$ cancel for every $\beta$, leaving $I=-1$.

References

• Cooper, Fred, Avinash Khare, and Uday Sukhatme. “Supersymmetry and Quantum Mechanics.” Physics Reports 251, no. 5–6 (1995): 267–385. doi:10.1016/0370-1573(94)00080-M. Open PDF.
• Weinberg, Steven. The Quantum Theory of Fields, Volume III: Supersymmetry. Cambridge: Cambridge University Press, 2000. doi:10.1017/CBO9781139644167.
• Witten, Edward. “Constraints on Supersymmetry Breaking.” Nuclear Physics B 202, no. 2 (1982): 253–316. doi:10.1016/0550-3213(82)90071-2.