Functional Determinants

⚡ Key Takeaways

The functional determinant $\det A$ of a (typically elliptic, self-adjoint) operator $A$ on a function space is a regularized infinite product of its eigenvalues, most often defined via the spectral zeta function as $\det A = e^{-\zeta_A'(0)}$ . It generalizes the determinant of a matrix to differential and integral operators, encodes one-loop quantum fluctuations, heat-kernel asymptotics, and analytic torsion. Developed by Ray–Singer (1971), Hawking, and Seeley; ubiquitous in quantum field theory, spectral geometry, and number theory.

💡 The Core Insight

A matrix determinant is $\det M = \prod_i \lambda_i = \exp(\operatorname{tr}\log M)$ . For an operator with infinitely many growing eigenvalues this product diverges, but the logarithm of the divergence is controlled: the function $s\mapsto \sum_n \lambda_n^{-s}$ converges for $\Re s$ large and continues meromorphically to a region containing $s=0$ , where it is regular. Since $\zeta_A(s) = \sum_n \lambda_n^{-s}$ formally has derivative $\zeta_A'(0) = -\sum_n \log\lambda_n$ , the quantity $\exp(-\zeta_A'(0))$ is the natural finite surrogate for $\prod_n \lambda_n$ . The regularity of $\zeta_A$ at $s=0$ is guaranteed by the heat-kernel expansion of an elliptic operator.

DefinitionSpectral Zeta Function and Zeta-Regularized Determinant

Let $A$ be a positive, self-adjoint, elliptic (pseudo-)differential operator of order $m>0$ on a compact manifold $M^d$ (possibly with boundary conditions), with discrete spectrum $0<\lambda_1\le\lambda_2\le\cdots\to\infty$ and finite-multiplicity eigenvalues. The spectral zeta function is

1

\quad \zeta_A(s) = \sum_{n=1}^\infty \lambda_n^{-s} = \frac{1}{\Gamma(s)}\int_0^\infty t^{s-1}\,\operatorname{Tr}\!\big(e^{-tA}\big)\,dt,\qquad \Re s > d/m.

Derivation hint: Use $\lambda^{-s}=\frac{1}{\Gamma(s)}\int_0^\infty t^{s-1}e^{-t\lambda}\,dt$ and sum over the spectrum, identifying $\operatorname{Tr}\,e^{-tA}=\sum_n e^{-t\lambda_n}$ . The functional (zeta-regularized) determinant is

2

\quad \det{}_\zeta A := \exp\!\big(-\zeta_A'(0)\big),\qquad \zeta_A'(0) = \frac{d}{ds}\Big|_{s=0}\zeta_A(s).

Derivation hint: Formally differentiate $\sum_n \lambda_n^{-s}$ in $s$ to get $-\sum_n \lambda_n^{-s}\log\lambda_n$ , evaluate at $s=0$ to recover $-\sum_n\log\lambda_n=-\log\prod_n\lambda_n$ , then exponentiate the negative. Here $\lambda_n^{-s} = e^{-s\log\lambda_n}$ with the principal branch of the logarithm.

TheoremRegularity at the Origin (Seeley)

For $A$ as above, $\operatorname{Tr}(e^{-tA})$ admits a small- $t$ asymptotic expansion

3

\quad \operatorname{Tr}\big(e^{-tA}\big)\ \sim\ \sum_{k=0}^{\infty} a_k\, t^{(k-d)/m} \;+\; \sum_{\ell=1}^\infty b_\ell\, t^{\ell}\log t \quad (t\to 0^+),

Derivation hint: This is the Seeley–DeWitt / heat-kernel expansion; the coefficients $a_k$ are integrals of local invariants built from the symbol of $A$ (curvature, potential, etc.). The $\log t$ terms generically vanish for differential operators. so that $\zeta_A(s)$ extends meromorphically to $\mathbb{C}$ with only simple poles at $s=(d-k)/m$ , and is holomorphic at $s=0$ . Consequently $\zeta_A'(0)$ and $\det_\zeta A$ are well defined. Moreover the value $\zeta_A(0)$ equals the constant ( $t^0$ ) heat coefficient minus the dimension of the kernel:

4

\quad \zeta_A(0) = a_{d} - \dim\ker A,

where $a_d$ is the coefficient of $t^0$ in expansion [3]. Derivation hint: Split the Mellin integral [1] at $t=1$ . The $\int_1^\infty$ piece is entire; the $\int_0^1$ piece, fed by [3], produces poles of $\zeta_A(s)\Gamma(s)$ . Since $1/\Gamma(s)\sim s$ near $s=0$ , the simple pole of $\Gamma$ at $0$ cancels, leaving a finite value at $s=0$ governed by the $t^0$ coefficient.

ProofFrom Heat Kernel to Determinant

Step 1 — Mellin representation. Starting from $\lambda^{-s}=\Gamma(s)^{-1}\int_0^\infty t^{s-1}e^{-t\lambda}dt$ , sum over eigenvalues (justified for $\Re s>d/m$ where both series and integral converge) to obtain [1].

Step 2 — Split the integral.

5

\quad \Gamma(s)\zeta_A(s) = \int_0^1 t^{s-1}\theta(t)\,dt + \int_1^\infty t^{s-1}\theta(t)\,dt,\qquad \theta(t):=\operatorname{Tr}e^{-tA}.

Derivation hint: The cutoff at $t=1$ is arbitrary; only the small- $t$ behavior governs analytic structure near $s=0$ .

Step 3 — Continue each piece. The second integral converges for all $s$ (since $\theta(t)\to \dim\ker A$ exponentially fast for the nonzero modes), giving an entire function. Insert the expansion [3] into the first integral:

6

\quad \int_0^1 t^{s-1}\,a_k\,t^{(k-d)/m}\,dt = \frac{a_k}{s + (k-d)/m},

Derivation hint: $\int_0^1 t^{s-1+\alpha}dt = 1/(s+\alpha)$ for $\Re(s+\alpha)>0$ , then continue meromorphically in $s$ ; each heat coefficient yields one simple pole. These simple poles at $s=(d-k)/m$ are the only singularities of $\Gamma(s)\zeta_A(s)$ .

Step 4 — Behavior at $s=0$ . Near $s=0$ , $\Gamma(s)=\frac1s-\gamma+O(s)$ . Writing $\Gamma(s)\zeta_A(s)=\frac{c_{-1}}{s}+c_0+\cdots$ with $c_{-1}=a_d-\dim\ker A$ (the $t^0$ residue) gives

7

\quad \zeta_A(s) = \big(s+\gamma s^2+\cdots\big)\Big(\frac{c_{-1}}{s}+c_0+c_1 s+\cdots\Big),

Derivation hint: Multiply $1/\Gamma(s)=s+\gamma s^2+\cdots$ by the Laurent series of $\Gamma\zeta_A$ ; collect powers of $s$ . so $\zeta_A(0)=c_{-1}$ (proving [4]) and $\zeta_A'(0)=c_0+\gamma c_{-1}$ , which is finite. Exponentiating $-\zeta_A'(0)$ yields the well-defined determinant [2]. $\blacksquare$

∎

mathematician · Daniel B. Ray & Isadore M. Singer | 1971

Ray and Singer introduced analytic torsion as a determinant-line invariant built from zeta-regularized determinants of the Laplacian on differential forms, conjecturing (later proven by Cheeger and Müller) its equality with the topological Reidemeister–Franz torsion. Their 1971 paper "R-torsion and the Laplacian on Riemannian manifolds" is the founding document of functional-determinant theory in geometry. Singer's broader program connected spectral data to topology (Atiyah–Singer index theorem); Robert Seeley supplied the analytic backbone by proving the meromorphic continuation and complex powers $A^{-s}$ of elliptic operators (1967).

🧠 Intuition

Think of an operator as an infinite diagonal matrix in its eigenbasis. The determinant "wants" to be the product of the diagonal. For a finite matrix this is $\exp(\sum\log\lambda_i)$ . For an operator the eigenvalues march off to infinity, so $\sum\log\lambda_n$ diverges linearly-ish. Zeta regularization is a principled way of throwing away the infinite part while keeping a canonical finite remainder — it is the analog of analytic continuation that gives $1+2+3+\cdots = -\tfrac1{12}$ via $\zeta(-1)$ . The key magic is that the divergence is local (it lives in the heat coefficients $a_k$ , which are integrals of curvature and potential), whereas the finite remainder $\zeta_A'(0)$ is global/spectral and carries genuine geometric information. So a functional determinant cleanly separates "ultraviolet junk" from "physical answer."

SH

Stephen Hawking

1977

In "Zeta function regularization of path integrals in curved spacetime" (1977), Hawking promoted $\det_\zeta A$ to a calculational tool for quantum field theory on curved backgrounds, computing one-loop effective actions and partition functions $Z=(\det A)^{-1/2}$ directly from spectral zeta functions. This method elegantly sidesteps the ambiguities of cutoff regularization, automatically respecting general covariance, and became the standard physicist's route to vacuum energy, the Casimir effect, and black-hole thermodynamics.

ExampleDeterminant of

-\partial_x^2

on an Interval (Dirichlet)

Given: $A=-\dfrac{d^2}{dx^2}$ on $[0,L]$ with Dirichlet boundary conditions. Eigenfunctions $\sin(n\pi x/L)$ , eigenvalues

\lambda_n=\left(\frac{n\pi}{L}\right)^2,\quad n=1,2,\dots

Step 1 — Spectral zeta.

\zeta_A(s)=\sum_{n=1}^\infty\left(\frac{n\pi}{L}\right)^{-2s}=\left(\frac{L}{\pi}\right)^{2s}\sum_{n=1}^\infty n^{-2s}=\left(\frac{L}{\pi}\right)^{2s}\zeta_R(2s),

where $\zeta_R$ is the Riemann zeta function.

Step 2 — Differentiate at $s=0$ .

\zeta_A'(s)=2\log\!\frac{L}{\pi}\left(\frac{L}{\pi}\right)^{2s}\zeta_R(2s)+\left(\frac{L}{\pi}\right)^{2s}\cdot 2\zeta_R'(2s).

Using $\zeta_R(0)=-\tfrac12$ and $\zeta_R'(0)=-\tfrac12\log(2\pi)$ :

\zeta_A'(0)=2\log\!\frac{L}{\pi}\cdot\big(-\tfrac12\big)+2\cdot\big(-\tfrac12\log 2\pi\big)=-\log\frac{L}{\pi}-\log 2\pi=-\log(2L).

Step 3 — Exponentiate.

\det{}_\zeta A=e^{-\zeta_A'(0)}=e^{\log(2L)}.

\boxed{\det{}_\zeta\!\left(-\partial_x^2\right)_{\text{Dir},[0,L]} = 2L}

(dimensionless after absorbing the length scale into the eigenvalue normalization; the result matches the Gel'fand–Yaglom formula).

ExampleOne-Loop Vacuum Energy via

\zeta_A(0)

Scaling

Given: A field with operator $A=-\partial_x^2+m^2$ on $[0,L]$ , Dirichlet; eigenvalues $\lambda_n=(n\pi/L)^2+m^2$ . We want how $\det_\zeta A$ rescales under $A\to\mu^2 A$ (changing units/regulator scale $\mu$ ).

Step 1 — Scaling identity. For $c>0$ ,

\zeta_{cA}(s)=\sum_n(c\lambda_n)^{-s}=c^{-s}\zeta_A(s).

Step 2 — Differentiate.

\zeta_{cA}'(0)=-\log c\,\zeta_A(0)+\zeta_A'(0).

Step 3 — Determinant scaling.

\det{}_\zeta(cA)=\exp\!\big(-\zeta_{cA}'(0)\big)=c^{\,\zeta_A(0)}\det{}_\zeta A.

For our operator the relevant $t^0$ heat coefficient gives, with $d=1$ , $m=2$ (operator order),

\zeta_A(0)=a_1-\dim\ker A = \frac{L\,? }{\cdots}\ \Rightarrow\ \zeta_A(0)=-\tfrac12-\tfrac{mL}{2\sqrt\pi}\cdot(\text{coeff}),

the essential point being:

\boxed{\det{}_\zeta(cA)=c^{\,\zeta_A(0)}\,\det{}_\zeta A}

Physical reading: Because $\zeta_A(0)\ne 0$ in general, the functional determinant — and hence the one-loop effective action $W=\tfrac12\log\det A$ — carries a scale dependence proportional to $\zeta_A(0)$ . This is precisely the conformal/scaling anomaly: the units (in inverse length squared) chosen for eigenvalues do not cancel, and $\zeta_A(0)$ is the anomaly coefficient.

Figure

Mellin split: convergence of

\zeta_A(s)

from the heat trace. Small-

t

region (gold) sources the poles; large-

t

region is entire.

```svg

```

⚠️ The Determinant Is Not Multiplicative

A frequent error is assuming $\det_\zeta(AB)=\det_\zeta A\cdot\det_\zeta B$ as for matrices. In general this fails. The discrepancy is the multiplicative anomaly:

8

\quad \log\frac{\det_\zeta(AB)}{\det_\zeta A\,\det_\zeta B}=F(A,B)\ne 0.

Derivation hint: For commuting positive elliptic operators, the anomaly $F(A,B)$ is computed by the Wodzicki residue of an expression built from $\log A$ and $\log B$ (Kontsevich–Vishik, Wodzicki); it vanishes when the orders/dimensions make the relevant residue zero. Also note $\det_\zeta A$ is not invariant under arbitrary similarity if the conjugating map is not appropriately bounded, and it generally depends on the choice of spectral cut (branch of $\log$ ).

CounterexampleScale Anomaly Breaks Naive Rescaling

Consider $A$ on a $d$ -manifold with $\zeta_A(0)\ne 0$ . Naively one might "set $\hbar=1$ " and expect $\det_\zeta(cA)=c^{\dim}\det_\zeta A$ with " $\dim=\infty$ ," which is meaningless, or expect scale-invariance $\det_\zeta(cA)=\det_\zeta A$ . Both are wrong: as derived in Example 2, $\det_\zeta(cA)=c^{\zeta_A(0)}\det_\zeta A$ . Concretely, for $-\partial_x^2$ on $[0,L]$ we have $\zeta_A(0)=\zeta_R(0)=-\tfrac12\ne 0$ , so rescaling eigenvalues by $c$ multiplies the determinant by $c^{-1/2}$ — a genuine, unavoidable scale dependence that is the seed of the trace anomaly in QFT.

connection

Functional determinants sit at a crossroads of analysis, geometry, and physics. They are computed in one dimension by the Gelfand-Yaglom Theorem, which replaces the infinite product by the solution of an initial-value problem. The required analytic continuation rests on the Heat Kernel Expansion and Seeley's theory of Complex Powers of Elliptic Operators. In topology they define Ray-Singer Analytic Torsion, equal to Reidemeister Torsion (Cheeger–Müller theorem). The multiplicative anomaly is governed by the Wodzicki Residue, the unique trace on pseudodifferential operators. In physics they form the one-loop term of the <a class="wiki-link" href="/wiki/effective-action">Effective Action</a> and underlie the <a class="wiki-link" href="/wiki/casimir-effect">Casimir Effect</a> and the Conformal Anomaly. Through index theory they connect to the Atiyah-Singer Index Theorem and the eta invariant.

🔬 Where this appears

⏳ History

- 1949 — Minakshisundaram–Pleijel: introduce the spectral zeta function

\sum\lambda_n^{-s}

for the Laplacian and prove its meromorphic continuation via heat-kernel asymptotics.

- 1967 — Seeley: rigorously constructs complex powers

A^{-s}

of elliptic pseudodifferential operators and the meromorphic structure of

\zeta_A

, the analytic foundation for

\det_\zeta

.

- 1971 — Ray & Singer: define analytic torsion using zeta-regularized determinants of form Laplacians; conjecture equality with Reidemeister torsion.

- 1976 — Dowker & Critchley: apply zeta regularization to effective potentials in curved spacetime, anticipating Hawking.

- 1977 — Hawking: systematizes zeta-function regularization of path integrals;

\det_\zeta

becomes a standard QFT tool.

- 1978 — Cheeger; 1978 — Müller: independently prove the Ray–Singer conjecture (Cheeger–Müller theorem).

- 1984 — Wodzicki: discovers the noncommutative residue (Wodzicki residue), the unique trace, later shown to control the multiplicative anomaly.

- 1985 — Bismut–Freed; Quillen: develop determinant line bundles, holonomy, and the curvature of the determinant — geometrizing

\det_\zeta

.

- 1994–95 — Kontsevich–Vishik; Friedlander: characterize the multiplicative anomaly and noncommutative analogues of determinants.

open-problem

Non-self-adjoint and complex spectra. For operators whose spectrum is not on the positive real axis, the choice of spectral cut (Agmon angle / ray of minimal growth) enters, and $\det_\zeta A$ can depend on it through the eta-type contributions; a fully canonical, cut-independent determinant does not always exist.

Determinant line bundles and anomalies. When eigenvalues cross zero in a family (e.g., spectral flow), $\det A$ vanishes and the natural object is a section of a determinant line bundle; its curvature, holonomy (Bismut–Freed), and global obstructions encode chiral and gravitational anomalies — making the "determinant" a global geometric quantity rather than a number.

Multiplicative anomaly in physics. Whether the Wodzicki-residue anomaly $F(A,B)$ has observable consequences (e.g., in dimensional-reduction or finite-temperature QFT) remains subtle and occasionally controversial; matching different regularization schemes requires careful tracking of $\zeta_A(0)$ and scale-setting.

Beyond elliptic operators. For hypoelliptic (sub-Laplacian), non-compact, or singular settings the heat expansion [3] may carry non-standard powers/logs, and the very holomorphy of $\zeta_A$ at $s=0$ can fail, leaving the determinant ill-defined or requiring renormalization choices.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also