---
title: 'Modified Canonical Energy: A Unified View'
url: https://www.emergentmind.com/topics/modified-canonical-energy
type: topic
---

# Modified Canonical Energy: A Unified View

Searching arXiv for the cited works and adjacent literature on modified canonical energy.
Modified canonical energy does not denote a single universally fixed object. In the literature represented here, it refers to several distinct but structurally related operations on the standard canonical framework: an Abbott–Deser–Tekin-based replacement for the Hollands–Wald canonical energy in covariant gravity, a pullback of canonical energy to Hertz-potential variables for Schwarzschild perturbations, a constrained reinterpretation of the origin of canonical Boltzmann weights in quantum statistical mechanics, an occupancy-based reinterpretation of the multiplier \(\beta\) in canonical thermodynamics, and the entropy-corrected fixed-\(N\) free-energy functional used in finite-temperature density-matrix perturbation theory [1603.02530], [1807.09883], [2603.12148], [2506.21650], [1503.07037]. A central source of ambiguity is therefore terminological: in some works the modification is genuinely definitional, in others it is representational, and in still others it concerns the entropy or thermodynamic potential rather than the energy observable itself.

## 1. Taxonomy of usages

Across these works, the phrase is best understood as a family resemblance rather than a single definition. The common theme is that a standard canonical object is retained at the abstract level while its derivation, representation, or thermodynamic interpretation is altered.

| Context | Central object | Nature of modification |
|---|---|---|
| Covariant gravity | \(\mathcal E(K;\delta_1\Psi,\delta_2\Psi)\) | ADT/Euler–Lagrange replacement for HW canonical energy |
| Schwarzschild stability | \(\mathscr E(\gamma_1,\gamma_2)\) | Pullback to Hertz-potential variables with manifest positivity |
| Quantum statistical mechanics | \(Z(\beta)=\mathrm{Tr}\,e^{-\beta\hat H}\) | Constrained origin via \(\delta(\hat C)\), not a new energy formula |
| Canonical thermodynamics | \(\mu_{\mathcal E}\), \(S=kH_{m,B}\) | Reinterpretation of \(\beta\) through occupancy entropy |
| Electronic-structure response | \(\Omega=E-T_eS\) | Canonical free energy under fixed-\(N\) perturbation theory |

The first two usages arise in gravitational stability theory and are closest to a literal modified canonical energy. The third explicitly denies that a new thermodynamic quantity has been introduced: the modification is the operator-theoretic framework from which canonical energy weights arise [2603.12148]. The fourth changes the entropy functional and thereby the temperature–multiplier relation, while leaving the Boltzmann one-particle weights and expected-energy formula intact [2506.21650]. The fifth concerns the finite-temperature canonical free energy, i.e. energy modified by an entropy term together with the fixed-particle-number constraint [1503.07037].

A recurring misconception is that “modified canonical energy” always means a deformed Hamiltonian or a non-Boltzmann canonical weight. The sources considered here do not support that reading uniformly. In particular, the constrained-quantum construction preserves \(\rho_\beta\propto e^{-\beta\hat H}\) and \(Z(\beta)=\mathrm{Tr}\,e^{-\beta\hat H}\) exactly, while the occupancy-based construction preserves \(\mu_{\mathcal E}=\frac{N}{Z}\sum_c \epsilon_c e^{-\beta\epsilon_c}\) but alters the interpretation of \(\beta\) [2603.12148], [2506.21650].

## 2. Euler–Lagrange and ADT modification in covariant gravity

A direct and explicit modified canonical energy was proposed for generally covariant systems by replacing the Hollands–Wald symplectic-current construction with an off-shell Abbott–Deser–Tekin current built from the Euler–Lagrange expressions alone [1603.02530]. The fields are \(\Psi=(g,\psi)\), and the defining current satisfies
\[
\sqrt{-g}\,\mathcal J^\mu
=
\partial_\nu\!\left(\sqrt{-g}\,\mathbf Q^{\mu\nu}\right),
\]
with
\[
\mathcal J^\mu(\zeta;\Psi,\delta\Psi)
=
\mathcal J^\mu_{\rm ADT}(\zeta;\Psi,\delta\Psi)
+
\mathcal J^\mu_{\Delta}(\Psi\mid \pounds_\zeta\Psi,\delta\Psi).
\]
For an exact Killing vector \(K\), the modified canonical energy is defined by
\[
\boxed{
\mathcal E(K;\delta_1\Psi,\delta_2\Psi)
\equiv
-\frac{1}{8\pi G}\int_\Sigma dx_\mu\,\sqrt{-g}\,
\mathcal J^\mu_{\rm ADT}(K;\delta_2\Psi,\delta_1\Psi)
}
\]
and, on-shell and bilinear in first-order perturbations,
\[
\boxed{
\mathcal E(K;\delta_1\Psi,\delta_2\Psi)
\equiv
-\frac1{8\pi G}\int_\Sigma dx_\mu\, \sqrt{-g}\,
(\delta_2\delta_1 \mathbf E^\mu{}_\nu)\,K^\nu
\Big|_{\delta\Psi=0,\ {\rm on\text{-}shell} }.
}
\]

The significance of this definition is precise. It depends only on the equations of motion and their linearizations, not on a chosen Lagrangian representative. This removes the ambiguity associated with shifts of the form \(L\to L+d\mu\), under which the presymplectic potential and symplectic current can change by boundary-exact terms. The modified construction therefore preserves covariance while decoupling canonical energy from Lagrangian boundary ambiguities.

Its formal properties parallel those of the Hollands–Wald energy. On-shell, the ADT bilinear form is symmetric,
\[
\left. \delta_2 \mathcal J^\mu_{\rm ADT}(\zeta;\Psi,\delta_1\Psi) \right|_{\rm on\text{-}shell}
=
\left. (\delta_2\delta_1 \mathbf E^\mu{}_\nu)\,\zeta^\nu \right|_{\rm on\text{-}shell}
=
\left. \delta_1 \mathcal J^\mu_{\rm ADT}(\zeta;\Psi,\delta_2\Psi) \right|_{\rm on\text{-}shell},
\]
is conserved for Killing \(K\), and is gauge invariant under compactly supported gauge transformations, with the noncompact black-hole case reduced to the same horizon and asymptotic conditions used by Hollands–Wald. The comparison with the original construction is especially sharp:
\[
\mathcal E_{\rm HW}(K;\delta\Psi,\delta\Psi)-\mathcal E(K;\delta\Psi,\delta\Psi)
=
\frac1{16\pi G} \int_{\partial\Sigma} dx_{\mu\nu}
\left[
2\sqrt{-g}\,\mathbf Q^{\mu\nu}(K;\delta\Psi,\delta\Psi)
-\sqrt{-g}\,\mathbf A^{\mu\nu}(\Psi\mid \pounds_K\delta\Psi,\delta\Psi)
\right]_{\rm on\text{-}shell}.
\]
Under the stated asymptotic conditions, the difference localizes to the bifurcation surface \(B\). This establishes that the modification is not arbitrary: it is a bulk-equivalent alternative differing from Hollands–Wald only by a horizon boundary term.

The same framework was applied to three-dimensional hairy extremal AdS black holes with
\[
L = R - \partial_\mu\varphi\,\partial^\mu\varphi - V(\varphi),
\]
for which the metric perturbation is pure gauge and the mixed metric–scalar terms vanish. The remaining scalar contribution reduces on the near-horizon geometry to
\[
\boxed{
\mathcal E(K;\delta\varphi,\delta\varphi)
=
\frac1{8\pi G} \int_\Sigma d\rho\,d\theta\, \frac{r_{NH}}{2\rho^2}
\left[
2\rho^2\,\delta\varphi^2
+\rho^4 (\partial_\rho\delta\varphi)^2
+(\partial_t\delta\varphi)^2
\right]
}
\]
which is manifestly nonnegative. In this usage, modified canonical energy is both a definition and a stability functional.

## 3. Hertz-potential reformulation and positivity on Schwarzschild

A second gravitational usage keeps the canonical energy itself unchanged but rewrites it on a different set of variables, thereby producing what the source explicitly describes as a practically modified expression [1807.09883]. The background is the \(4\)-dimensional Schwarzschild exterior,
\[
ds^2=\frac{\Delta}{r^2}dt^2-\frac{r^2}{\Delta}dr^2-r^2(d\theta^2+\sin^2\theta\,d\phi^2),
\qquad
\Delta=r(r-2M),
\]
and the canonical energy is the Hollands–Wald bilinear form
\[
\mathscr E(\gamma_1,\gamma_2)=\Omega(\gamma_1,\pounds_t\gamma_2),
\qquad
\mathscr E(\gamma):=\mathscr E(\gamma,\gamma).
\]
In ADM variables it admits the static split \(\mathscr E=\mathscr K+\mathscr U\), but manifest positivity is obstructed by the linearized constraints, the gauge dependence of the density, and degeneracy on pure gauge or stationary directions.

The reformulation proceeds through the Teukolsky–adjoint identity
\[
S\,E = O\,T,
\qquad
T^\dagger O^\dagger = E^\dagger S^\dagger,
\]
so that any solution of \(O^\dagger[\psi]=0\) generates a metric perturbation
\[
\gamma^{ab}=S^\dagger[\psi].
\]
On Schwarzschild, the perturbation in ingoing radiation gauge is
\[
\gamma_{ab} = -l_al_b\,U + l_{(a}m_{b)}\,V - m_am_b\,W,
\]
with
\[
U=\eth^2\psi,\qquad
V=\big[\thorn\eth+\eth(\thorn+3\rho)\big]\psi,\qquad
W=(\thorn-\rho)(\thorn+3\rho)\psi.
\]
The free Teukolsky initial datum is
\[
\eta := (\thorn+3\rho)\psi.
\]
Because \(\psi\) satisfies a scalar decoupled equation and \((\psi,\eta)|_\Sigma\) are free initial data, the resulting metric perturbation automatically solves the linearized Einstein equation and is already in a fixed radiation gauge. The practical modification is therefore not a new bilinear form but the elimination of the original constraint and gauge obstructions.

The central formula is the pullback of canonical energy to Hertz-potential variables:
\[
\boxed{
\mathscr E = \frac{1}{64\pi} \int_\Sigma N
\Big[
|(\thorn+\rho)V|^2
+
|D_tW-\eth'V|^2
+
|D_rW+\eth'V|^2
\Big].
}
\]
This is manifestly nonnegative. The source proves positivity first for the complex Hertz-generated perturbation and then, using
\[
\Omega(\gamma,\gamma')=0
\qquad
(\gamma=S^\dagger[\psi],\ \gamma'=S^\dagger[\psi']),
\]
deduces positivity for the associated real perturbation:
\[
\mathscr E(\Re\gamma)=\frac12 \mathscr E(\gamma)\ge 0.
\]

The same formalism identifies the Dafermos–Holzegel–Rodnianski energy as canonical energy of an associated Hertz-generated perturbation. Defining
\[
\psi=r^4\hat\psi_4,
\qquad
\Psi_{DHR}=(\thorn+\rho)(\thorn+3\rho)\psi,
\]
the final relation is
\[
\boxed{
\mathscr E[\gamma]=\frac{1}{4\pi}E_{DHR}[\hat\gamma].
}
\]
Thus the DHR energy is not an unrelated substitute but an equivalent stability energy obtained after a Teukolsky/Hertz transform. In this context, “modified canonical energy” denotes a variable change that makes positivity manifest without changing the abstract definition \(\mathscr E(\gamma_1,\gamma_2)=\Omega(\gamma_1,\pounds_t\gamma_2)\).

## 4. Constrained quantum origin of canonical and microcanonical statistics

In quantum statistical mechanics, the relevant work explicitly states that it does not introduce a new thermodynamic quantity called modified canonical energy in the sense of altering the Hamiltonian energy entering the canonical ensemble [2603.12148]. Its contribution is structural: it reformulates the origin of both canonical and microcanonical statistical descriptions inside a constrained quantum framework.

The construction enlarges the Hilbert space to
\[
\mathcal H_{\mathrm{ext}}=\mathcal H_T\otimes \mathcal H_{\mathrm{QM}},
\]
with auxiliary clock operators satisfying
\[
[\hat T,\hat P_T]=i,
\qquad
[\hat T,\hat H]=[\hat P_T,\hat H]=0.
\]
The constraint operator is
\[
\hat C=\hat P_T+\hat H,
\qquad
\hat C\,|\Psi_{\mathrm{phys}}\rangle=0,
\]
and the unifying object is the Dirac projector
\[
\delta(\hat C)
=
\frac{1}{2\pi}\int_{-\infty}^{+\infty} d\alpha\,e^{i\alpha \hat C}
=
\frac{1}{2\pi}\int_{-\infty}^{+\infty} d\alpha\,e^{i\alpha(\hat P_T+\hat H)}.
\]

In the clock-time basis \(|T,q\rangle=|T\rangle\otimes|q\rangle\), the physical kernel is
\[
K_{\mathrm{phys}}(T_f,q_f;T_i,q_i)
=
\langle T_f,q_f|\delta(\hat C)|T_i,q_i\rangle
=
\langle q_f|e^{i(T_i-T_f)\hat H}|q_i\rangle.
\]
A purely imaginary clock separation,
\[
T_f-T_i=-i\beta,
\]
yields the Euclidean kernel
\[
K_\beta(q_f,q_i)=\langle q_f|e^{-\beta\hat H}|q_i\rangle,
\]
and tracing gives the canonical partition function
\[
Z(\beta)=\mathrm{Tr}\,e^{-\beta \hat H}.
\]

In the conjugate clock-energy basis, with
\[
\hat P_T|E\rangle=-E|E\rangle,
\qquad
\langle E_f|E_i\rangle=\delta(E_f-E_i),
\]
the same projector reduces to
\[
K(E_f,q_f;E_i,q_i)
=
\delta(E_f-E_i)\langle q_f|\delta(\hat H-E_i)|q_i\rangle,
\]
so that
\[
\Omega(E)=\mathrm{Tr}\,\delta(\hat H-E).
\]
The result is the paper’s central structural claim: canonical and microcanonical ensembles emerge as complementary projections in different clock bases.

The conceptual consequence is limited but important. What changes is not the canonical formula
\[
\rho_\beta \propto e^{-\beta \hat H},
\qquad
Z(\beta)=\mathrm{Tr}\,e^{-\beta \hat H},
\]
but the status of canonical energy within the theory. Energy becomes relational, enforced by the constraint \(\hat P_T+\hat H=0\), so that the system Hamiltonian is one component of a reparametrization-invariant structure rather than an isolated primitive. This suggests that canonical and microcanonical statistics need not be introduced as independent constructions. It does not, however, derive the gravitational canonical energy of black-hole perturbation theory or a new covariant thermodynamic first law.

## 5. Occupancy-number thermodynamics and reinterpretation of \(\beta\)

A different usage arises in canonical thermodynamics derived from the multinomial distribution of occupancy numbers [2506.21650]. Here the paper does not replace the Boltzmann form
\[
P_c=Z^{-1}e^{-\beta \epsilon_c},
\qquad
Z=\sum_{c\in\mathbb C} e^{-\beta\epsilon_c},
\]
nor the expected-energy formula
\[
\mu_{\mathcal E} = \frac{N}{Z}\sum_{c\in\mathbb C}\epsilon_c e^{-\beta \epsilon_c}.
\]
Instead, it changes the entropy assigned to the canonical system and thereby changes the interpretation of \(\beta\).

The system is a closed \(N\)-particle system with i.i.d. one-particle eigenstates. If \(n_c\) is the occupancy number of state \(c\), the occupancy vector \(\bar n\) is multinomial,
\[
P_{\bar n}=W_{\bar n}\prod_{c\in\mathbb C} P_c^{\,n_c},
\qquad
W_{\bar n}=\frac{N!}{\prod_{c\in\mathbb C} n_c!}.
\]
Its Shannon entropy is
\[
H_m=-E\{\log P_{\bar n}\}
=
N H_c-\log(N!)+\sum_{c\in\mathbb C}E\{\log(n_c!)\},
\]
where
\[
H_c=-\sum_{c\in\mathbb C}P_c\log P_c.
\]
Because the exact entropy-maximizing categorical law appears analytically intractable, the Boltzmann one-particle law is adopted as an approximation, producing the multinomial-Boltzmann distribution
\[
P_{\bar n} = W_{\bar n} Z^{-N}\prod_{c\in\mathbb C} e^{-\beta \epsilon_c n_c}.
\]

The resulting thermodynamic entropy is identified as
\[
S=kH_{m,B},
\]
with
\[
H_{m,B}
=
\beta \mu_{\mathcal E}+N\log Z - \log(N!)
+\sum_{c\in\mathbb C} E\{\log(n_c!)\}.
\]
Imposing Clausius’ relation,
\[
\frac{dS}{d\mu_{\mathcal E}}=\frac{1}{T},
\]
gives the key equation
\[
\frac{dS}{k\, d\mu_{\mathcal E}}
=
\beta + \sum_{c\in\mathbb C} \frac{d\,E\{\log(n_c!)\}}{d\mu_{\mathcal E}}
=
\frac{1}{kT},
\]
hence
\[
\boxed{
\beta = \frac{1}{kT}
-
\sum_{c\in\mathbb C}\frac{d\,E\{\log(n_c!)\}}{d\mu_{\mathcal E}}.
}
\]
The source therefore concludes that, in general,
\[
\beta \neq \frac{1}{kT}
\]
outside the one-particle and classical dilute limits.

This is not a deformation of the expected energy. The modification lies in the entropy functional and the temperature–energy relation. The source uses a Bose–Einstein comparison as supporting evidence: if one forces \(\beta=1/(kT)\) in the multinomial-Boltzmann occupancy theory, the entropy as a function of temperature can exceed the Bose–Einstein thermodynamic entropy at the same temperature, which is treated as incompatible with the maximum entropy principle. The claim is explicitly strongest for noninteracting quantum systems, relies on the i.i.d. assumption, and is supported by numerical evidence rather than a general theorem. In this usage, modified canonical energy is essentially a shorthand for a modified canonical thermodynamic interpretation.

## 6. Canonical free energy in finite-temperature density-matrix perturbation theory

In finite-temperature electronic-structure theory, the exact phrase modified canonical energy is not used, but the closest object is the canonical free energy: energy modified by an entropy term and by the fixed-\(N\) constraint [1503.07037]. The setting is tight-binding, Hartree–Fock, and Kohn–Sham DFT at finite electronic temperature \(T_e\).

For the non-self-consistent orthogonal one-particle case, the thermodynamic functional is
\[
\Omega(\lambda)
=
\mathrm{Tr}[P(\lambda)H(\lambda)]
-
T_e \,\mathcal{S}[P(\lambda)],
\]
with
\[
\mathcal{S}[P]
=
-k_B \mathrm{Tr}[P\ln P + (I-P)\ln(I-P)].
\]
For the self-consistent case,
\[
\Omega_{\rm SCF}[D]
=
2\mathrm{Tr}[hD] + \mathrm{Tr}[DG(D)] - 2T_e\,\mathcal{S}[D^\perp],
\]
subject to
\[
2\mathrm{Tr}[DS] = N_e.
\]
The finite-temperature density matrix is
\[
P = \left[e^{\beta(H-\mu I)} + I\right]^{-1},
\qquad
\beta = \frac{1}{k_B T_e},
\]
and the defining canonical feature is that \(\mu\) is not externally fixed. Instead it is expanded order by order,
\[
\mu(\lambda)=\mu^{(0)}+\lambda \mu^{(1)}+\lambda^2 \mu^{(2)}+\cdots,
\]
so that
\[
\mathrm{Tr}[P^{(0)}]=N_{\rm occ},
\qquad
\mathrm{Tr}[P^{(k)}]=0,\quad k>0.
\]

The perturbation theory is implemented by a recursive Fermi-operator expansion. With
\[
X_0 = \frac{1}{2}I - 2^{-(M+2)}\beta (H-\mu I),
\qquad
X_n = \frac{X_{n-1}^2}{X_{n-1}^2 + (I-X_{n-1})^2},
\]
the final density matrix is \(P=X_M\). Perturbative coefficients are propagated through the recursion, and the chemical-potential correction uses
\[
P_\mu = \beta P^{(0)}(I-P^{(0)}),
\]
together with the Newton updates
\[
\mu^{(0)} \leftarrow \mu^{(0)} + \frac{N_e-\mathrm{Tr}[P^{(0)}]}{\mathrm{Tr}[P_\mu]},
\qquad
\mu^{(i)} \leftarrow \mu^{(i)} - \frac{\mathrm{Tr}[P^{(i)}]}{\mathrm{Tr}[P_\mu]}.
\]

The principal response formula for the non-self-consistent free energy is
\[
\boxed{
\Omega^{(m)} = \frac{1}{m}\sum_{k=1}^{m} k\,\mathrm{Tr}\!\left[H^{(k)}P^{(m-k)}\right].
}
\]
For self-consistent Hartree–Fock or Kohn–Sham theory, the corresponding result is
\[
\boxed{
\Omega_{\rm SCF}^{(m)} = \frac{2}{m}\mathrm{Tr}[h^{(1)}D^{(m-1)}],
\qquad m>0.
}
\]
These formulas are the finite-\(T\) fixed-\(N\) analogues of density-matrix response theory. The operative energetic quantity is therefore not bare band energy but the canonical free energy \(E-T_eS\), evaluated on a manifold where particle number is held fixed by perturbative adjustment of \(\mu\).

This suggests a narrower terminological conclusion. In electronic-structure response theory, “modified canonical energy” is most precisely interpreted as the canonical free energy itself, together with its perturbative derivatives, rather than as a new energy observable. The modification comes from finite-\(T\) entropy and canonical number projection.

## 7. Unifying perspective and limits of the term

The surveyed constructions show that the modification can occur at four different levels. It can be **definitional**, as in the ADT/Euler–Lagrange replacement for Hollands–Wald canonical energy [1603.02530]. It can be **representational**, as in the Hertz-potential pullback that yields a manifestly positive quadratic form on Schwarzschild perturbations [1807.09883]. It can be **operator-theoretic**, as in the constrained projector \(\delta(\hat C)\) from which canonical and microcanonical ensembles emerge as complementary clock-sector projections [2603.12148]. Or it can be **thermodynamic**, through a revised entropy functional and hence a revised \(\beta\)-temperature relation [2506.21650], or through the finite-\(T\) free energy \(E-T_eS\) under fixed particle number [1503.07037].

The term therefore has no single invariant content across subfields. In gravitational stability, it usually denotes a bona fide alteration of the canonical-energy functional or of its variable representation. In equilibrium statistical mechanics and finite-temperature many-electron theory, it more often denotes a reinterpretation of the origin or thermodynamic role of canonical quantities while retaining standard canonical weights or energies. A precise reading accordingly requires specifying whether the modification acts on the functional definition, the choice of variables, the constraint structure, the entropy, or the thermodynamic potential.

Source: https://www.emergentmind.com/topics/modified-canonical-energy