Functional Determinants

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 $det A = e^{- ζ_{A}^{'} (0)}$ . 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.

For a positive-definite operator $A$ with discrete spectrum ${λ_{n}}$ , the naive product $\prod_{n} λ_{n}$ diverges. The functional determinant assigns a finite value by zeta-function regularization: one forms the spectral zeta function $ζ_{A} (s) = \sum_{n} λ_{n}^{- s}$ , analytically continues it to $s = 0$ , and defines $det A = exp (- ζ_{A}^{'} (0))$ . This object matters because the Gaussian functional integral $\int D ϕ e^{- \frac{1}{2} ⟨ ϕ, A ϕ ⟩}$ formally equals $(det A)^{- 1/2}$ , making functional determinants the engine of one-loop effective actions. A surprising fact: the regularized product is not multiplicative in general — $det (A B) \neq = det A \cdot det B$ — the failure being measured by the multiplicative anomaly (Wodzicki residue).

The Core Insight

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

Formal Statement

Spectral 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 $0 < λ_{1} \leq λ_{2} \leq \dots \to \infty$ and finite-multiplicity eigenvalues. The spectral zeta function is

ζ_{A} (s) = n = 1 \sum \infty λ_{n}^{- s} = \frac{1}{Γ ( s )} \int_{0}^{\infty} t^{s - 1} Tr (e^{- t A}) d t, ℜ s > d / m .

⟳Use

λ^{- s} = \frac{1}{Γ ( s )} \int_{0}^{\infty} t^{s - 1} e^{- t λ} d t

and sum over the spectrum, identifying

Tr e^{- t A} = \sum_{n} e^{- t λ_{n}}

The functional (zeta-regularized) determinant is

det_{ζ} A := exp (- ζ_{A}^{'} (0)), ζ_{A}^{'} (0) = \frac{d}{d s}_{s = 0} ζ_{A} (s) .

⟳Formally differentiate

\sum_{n} λ_{n}^{- s}

s

to get

- \sum_{n} λ_{n}^{- s} lo g λ_{n}

, evaluate at

s = 0

to recover

- \sum_{n} lo g λ_{n} = - lo g \prod_{n} λ_{n}

, then exponentiate the negative.

Here $λ_{n}^{- s} = e^{- s l o g λ_{n}}$ with the principal branch of the logarithm.

Regularity at the Origin (Seeley)

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

Tr (e^{- t A}) \sim k = 0 \sum \infty a_{k} t^{(k - d) / m} + ℓ = 1 \sum \infty b_{ℓ} t^{ℓ} lo g t (t \to 0^{+}),

⟳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

lo g t

terms generically vanish for differential operators.

so that $ζ_{A} (s)$ extends meromorphically to $C$ with only simple poles at $s = (d - k) / m$ , and is holomorphic at $s = 0$ . Consequently $ζ_{A}^{'} (0)$ and $det_{ζ} A$ are well defined. Moreover the value $ζ_{A} (0)$ equals the constant ( $t^{0}$ ) heat coefficient minus the dimension of the kernel:

ζ_{A} (0) = a_{d} - dim ker A,

where $a_{d}$ is the coefficient of $t^{0}$ in expansion [3].

⟳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

ζ_{A} (s) Γ (s)

. Since

1/Γ (s) \sim s

near

s = 0

, the simple pole of

Γ

0

cancels, leaving a finite value at

s = 0

governed by the

t^{0}

coefficient.

Why It Holds / Derivation

From Heat Kernel to Determinant

Step 1 — Mellin representation. Starting from

λ^{- s} = Γ (s)^{- 1} \int_{0}^{\infty} t^{s - 1} e^{- t λ} d t

, sum over eigenvalues (justified for

ℜ s > d / m

where both series and integral converge) to obtain [1]. Step 2 — Split the integral.

Γ (s) ζ_{A} (s) = \int_{0}^{1} t^{s - 1} θ (t) d t + \int_{1}^{\infty} t^{s - 1} θ (t) d t, θ (t) := Tr e^{- t A} .

⟳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

θ (t) \to dim ker A

exponentially fast for the nonzero modes), giving an entire function. Insert the expansion [3] into the first integral:

\int_{0}^{1} t^{s - 1} a_{k} t^{(k - d) / m} d t = \frac{a _{k}}{s + ( k - d ) / m},

⟳

\int_{0}^{1} t^{s - 1 + α} d t = 1/ (s + α)

for

ℜ (s + α) > 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 $Γ (s) ζ_{A} (s)$ .

Step 4 — Behavior at $s = 0$ . Near

s = 0

Γ (s) = \frac{1}{s} - γ + O (s)

. Writing

Γ (s) ζ_{A} (s) = \frac{c _{- 1}}{s} + c_{0} + \dots

with

c_{- 1} = a_{d} - dim ker A

(the

t^{0}

residue) gives

ζ_{A} (s) = (s + γ s^{2} + \dots) (\frac{c _{- 1}}{s} + c_{0} + c_{1} s + \dots),

⟳Multiply

1/Γ (s) = s + γ s^{2} + \dots

by the Laurent series of

Γ ζ_{A}

; collect powers of

s

so $ζ_{A} (0) = c_{- 1}$ (proving [4]) and $ζ_{A}^{'} (0) = c_{0} + γ c_{- 1}$ , which is finite. Exponentiating $- ζ_{A}^{'} (0)$ yields the well-defined determinant [2]. $■$

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).

Building 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 $exp (\sum lo g λ_{i})$ . For an operator the eigenvalues march off to infinity, so $\sum lo g λ_{n}$ 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 $1 + 2 + 3 + \dots = - \frac{1}{12}$ via $ζ (- 1)$ . 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 $ζ_{A}^{'} (0)$ is global/spectral and carries genuine geometric information. So a functional determinant cleanly separates "ultraviolet junk" from "physical answer."

Stephen Hawking | 1977

In "Zeta function regularization of path integrals in curved spacetime" (1977), Hawking promoted $det_{ζ} A$ 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.

Worked Examples

Determinant of $-\partial_x^2$ on an Interval (Dirichlet)

Given:

A = - \frac{d ^{2}}{d x ^{2}}

[0, L]

with Dirichlet boundary conditions. Eigenfunctions

sin (nπ x / L)

, eigenvalues

λ_{n} = (\frac{nπ}{L})^{2}, n = 1, 2, \dots

Step 1 — Spectral zeta.

ζ_{A} (s) = n = 1 \sum \infty (\frac{nπ}{L})^{- 2 s} = (\frac{L}{π})^{2 s} n = 1 \sum \infty n^{- 2 s} = (\frac{L}{π})^{2 s} ζ_{R} (2 s),

where

ζ_{R}

is the Riemann zeta function. Step 2 — Differentiate at $s = 0$ .

ζ_{A}^{'} (s) = 2 lo g \frac{L}{π} (\frac{L}{π})^{2 s} ζ_{R} (2 s) + (\frac{L}{π})^{2 s} \cdot 2 ζ_{R}^{'} (2 s) .

Using

ζ_{R} (0) = - \frac{1}{2}

and

ζ_{R}^{'} (0) = - \frac{1}{2} lo g (2 π)

ζ_{A}^{'} (0) = 2 lo g \frac{L}{π} \cdot (- \frac{1}{2}) + 2 \cdot (- \frac{1}{2} lo g 2 π) = - lo g \frac{L}{π} - lo g 2 π = - lo g (2 L) .

Step 3 — Exponentiate.

det_{ζ} A = e^{- ζ_{A}^{'} (0)} = e^{l o g (2 L)} .

det_{ζ} (- \partial_{x}^{2})_{Dir, [0, L]} = 2 L

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

One-Loop Vacuum Energy via $\zeta_A(0)$ Scaling

Given: A field with operator

A = - \partial_{x}^{2} + m^{2}

[0, L]

, Dirichlet; eigenvalues

λ_{n} = (nπ / L)^{2} + m^{2}

. We want how

det_{ζ} A

rescales under

A \to μ^{2} A

(changing units/regulator scale

μ

). Step 1 — Scaling identity. For

c > 0

ζ_{c A} (s) = n \sum (c λ_{n})^{- s} = c^{- s} ζ_{A} (s) .

Step 2 — Differentiate.

ζ_{c A}^{'} (0) = - lo g c ζ_{A} (0) + ζ_{A}^{'} (0) .

Step 3 — Determinant scaling.

det_{ζ} (c A) = exp (- ζ_{c A}^{'} (0)) = c^{ζ_{A} (0)} det_{ζ} A .

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

ζ_{A} (0) = a_{1} - dim ker A = \frac{L ?}{\dots} \Rightarrow ζ_{A} (0) = - \frac{1}{2} - \frac{m L}{2 π} \cdot (coeff),

the essential point being:

det_{ζ} (c A) = c^{ζ_{A} (0)} det_{ζ} A

Physical reading: Because

ζ_{A} (0) \neq = 0

in general, the functional determinant — and hence the one-loop effective action

W = \frac{1}{2} lo g det A

— carries a scale dependence proportional to

ζ_{A} (0)

. This is precisely the conformal/scaling anomaly: the units (in inverse length squared) chosen for eigenvalues do not cancel, and

ζ_{A} (0)

is the anomaly coefficient.

Mellin split: convergence of $\zeta_A(s)$ from the heat trace. Small-$t$ region (gold) sources the poles; large-$t$ region is entire.

<svg viewBox="0 0 640 360" xmlns="http://www.w3.org/2000/svg">
<rect width="640" height="360" fill="#0a0e17"/>
<line x1="60" y1="300" x2="600" y2="300" stroke="#94a3b8" stroke-width="1.5"/>
<line x1="60" y1="40" x2="60" y2="300" stroke="#94a3b8" stroke-width="1.5"/>
<text x="610" y="305" fill="#e2e8f0" font-size="14">t</text>
<text x="30" y="50" fill="#e2e8f0" font-size="14">θ(t)</text>
<text x="40" y="304" fill="#e2e8f0" font-size="12">0</text>
<!-- cutoff line t=1 -->
<line x1="300" y1="40" x2="300" y2="300" stroke="#94a3b8" stroke-width="1" stroke-dasharray="5 4"/>
<text x="288" y="320" fill="#e2e8f0" font-size="12">t=1</text>
<!-- heat trace curve: diverges like t^{-d/m} at 0, decays to dim ker at infinity -->
<path d="M70,60 C120,150 170,235 300,262 C420,283 520,288 595,289" fill="none" stroke="#d4a853" stroke-width="2.5"/>
<text x="360" y="250" fill="#d4a853" font-size="13">θ(t)=Tr e^{-tA}</text>
<!-- asymptote to dim ker -->
<line x1="60" y1="289" x2="600" y2="289" stroke="#94a3b8" stroke-width="0.8" stroke-dasharray="2 4"/>
<text x="500" y="284" fill="#94a3b8" font-size="11">→ dim ker A</text>
<!-- shade small-t region -->
<rect x="60" y="40" width="240" height="260" fill="#d4a853" fill-opacity="0.08"/>
<text x="90" y="100" fill="#d4a853" font-size="12">small t:</text>
<text x="90" y="118" fill="#d4a853" font-size="12">Σ a_k t^{(k-d)/m}</text>
<text x="90" y="136" fill="#d4a853" font-size="11">→ poles of ζ_A(s)</text>
<!-- shade large-t region -->
<rect x="300" y="40" width="300" height="260" fill="#94a3b8" fill-opacity="0.05"/>
<text x="380" y="100" fill="#e2e8f0" font-size="12">large t:</text>
<text x="380" y="118" fill="#e2e8f0" font-size="11">∫₁^∞ entire in s</text>
</svg>

Common Misconceptions

The Determinant Is Not Multiplicative

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

lo g \frac{det _{ζ} ( A B )}{det _{ζ} A det _{ζ} B} = F (A, B) \neq = 0.

⟳For commuting positive elliptic operators, the anomaly

F (A, B)

is computed by the Wodzicki residue of an expression built from

lo g A

and

lo g B

(Kontsevich–Vishik, Wodzicki); it vanishes when the orders/dimensions make the relevant residue zero.

Also note

det_{ζ} 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

lo g

Scale Anomaly Breaks Naive Rescaling

Consider $A$ on a $d$ -manifold with $ζ_{A} (0) \neq = 0$ . Naively one might "set $ℏ = 1$ " and expect $det_{ζ} (c A) = c^{d i m} det_{ζ} A$ with " $dim = \infty$ ," which is meaningless, or expect scale-invariance $det_{ζ} (c A) = det_{ζ} A$ . Both are wrong: as derived in Example 2, $det_{ζ} (c A) = c^{ζ_{A} (0)} det_{ζ} A$ . Concretely, for $- \partial_{x}^{2}$ on $[0, L]$ we have $ζ_{A} (0) = ζ_{R} (0) = - \frac{1}{2} \neq = 0$ , 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.

Connections

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 Effective Action and underlie the Casimir Effect and the Conformal Anomaly. Through index theory they connect to the Atiyah-Singer Index Theorem and the eta invariant.

Real-World Applications

Casimir energy — The attractive force between parallel plates is $E = \frac{1}{2} \sum_{n} ω_{n}$ , regularized as $- \frac{1}{2} ζ_{A}^{'} (- 1)$ for the relevant Laplacian; gives the celebrated $- π^{2} ℏ c / (720 d^{3})$ pressure, confirmed experimentally (Lamoreaux 1997).
Black-hole entropy and partition functions — Hawking's one-loop $Z = (det A)^{- 1/2}$ on Euclidean Schwarzschild/de Sitter backgrounds yields semiclassical free energies and entropy corrections.
Lattice QCD / instanton physics — The 't Hooft instanton determinant, a functional determinant of fluctuations around a self-dual gauge field, controls tunneling amplitudes and the $θ$ -vacuum.
Polyakov string measure — The bosonic string path integral produces $det^{'} Δ$ on Riemann surfaces; its dependence on the conformal factor (Liouville action) is fixed by $ζ_{Δ} (0)$ and the central charge $c = 26$ .
Analytic torsion in topology — Computes Reidemeister torsion of lens spaces, distinguishing homotopy-equivalent but non-homeomorphic manifolds.
Spectral statistics & number theory — $det_{ζ} (- Δ + const)$ relates to special values of Selberg and Dedekind zeta functions; the determinant of the Laplacian on a torus involves the Dedekind $η$ function.
Condensed matter / stochastic processes — Fluctuation determinants compute escape rates (Kramers), nucleation prefactors, and partition functions of fluctuating membranes.

Historical Development

1949 — Minakshisundaram–Pleijel: introduce the spectral zeta function $\sum λ_{n}^{- s}$ 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 $ζ_{A}$ , the analytic foundation for $det_{ζ}$ .
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; $det_{ζ}$ 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 $det_{ζ}$ .
1994–95 — Kontsevich–Vishik; Friedlander: characterize the multiplicative anomaly and noncommutative analogues of determinants.

Subtleties & Limits

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_{ζ} 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

ζ_{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

ζ_{A}

s = 0

can fail, leaving the determinant ill-defined or requiring renormalization choices.