Loschmidt Echo and Dynamical Quantum Phase Transitions in the TFIM
Setting
After preparing the system in the ground state $|\psi_0\rangle$ of an "initial" Hamiltonian $H_0 = -J\sum_i \sigma^z_i \sigma^z_{i+1} - h_0 \sum_i \sigma^x_i$ and suddenly quenching the field to $h_f$, the time-evolved state $|\psi(t)\rangle = e^{-i H_f t}|\psi_0\rangle$ has overlap with the initial state given by the Loschmidt amplitude
\[G(t) \;=\; \langle \psi_0 | e^{-i H_f t} | \psi_0 \rangle.\]
The Loschmidt echo is its squared modulus and the rate function (a quench dynamical free-energy density) is
\[L(t) \;=\; |G(t)|^2, \qquad \lambda(t) \;=\; -\frac{1}{N} \log L(t).\]
Heyl, Polkovnikov, and Kehrein observed [Heyl2013] that $\lambda(t)$ exhibits non-analytic cusps at a discrete set of critical times $t_n^*$ whenever the quench crosses the equilibrium quantum critical point — these are dynamical quantum phase transitions (DQPTs). The TFIM is the canonical solvable example; the review [Heyl2018] traces every subsequent generalization back to this case.
Per-mode product structure
Both $H_0$ and $H_f$ share the Jordan–Wigner / Bogoliubov decoupling into independent fermionic modes. Mode $n$ in $H_0$ rotates onto mode $n$ in $H_f$ by a Bogoliubov angle difference $\Delta\theta_n = \theta_n^{(0)} - \theta_n^{(f)}$, and only this two-dimensional subspace contributes to the per-mode amplitude. The full echo factorizes over modes,
\[\boxed{\; L(t) \;=\; \prod_n \Bigl| \cos^2\!\Delta\theta_n + \sin^2\!\Delta\theta_n \, e^{-2 i \Lambda_n^{(f)} t} \Bigr|^2, \;}\]
with $\Lambda_n^{(f)}$ the H_f single-quasiparticle energy of mode $n$.
Infinite (continuous-$k$) form
In the thermodynamic limit, momentum becomes continuous and the product turns into an integral. With
\[\Lambda_k(h) = 2\sqrt{J^2 + h^2 - 2 J h \cos k}, \qquad \tan(2\theta_k(h)) = \frac{J \sin k}{h - J \cos k},\]
the rate function reads [^Heyl2013, eq. (3)]
\[\boxed{\; \lambda(t) \;=\; -\frac{1}{2\pi}\int_0^\pi \log\!\Bigl| \cos^2\!\Delta\theta_k + \sin^2\!\Delta\theta_k \, e^{-2 i \Lambda_k(h_f) t} \Bigr|^2 \, dk. \;}\]
QAtlas evaluates this integral with QuadGK.quadgk directly; the DQPT cusp is an integrable logarithmic singularity that adaptive Gauss–Kronrod handles without special bookkeeping.
DQPT critical times
Set $\alpha_k = \cos^2\Delta\theta_k$, $\beta_k = \sin^2\Delta\theta_k$. The integrand $|\alpha_k + \beta_k e^{-2 i \Lambda_f t}|^2$ vanishes iff $\alpha_k = \beta_k$ and $e^{-2 i \Lambda_f t} = -1$, i.e.
\[\cos(2\Delta\theta_{k^*}) = 0, \qquad 2 \Lambda_{k^*}^{(f)} t_n^* = (2n+1)\pi.\]
So the DQPT critical times are
\[\boxed{\; t_n^* \;=\; \frac{\pi (n + \tfrac{1}{2})}{\Lambda_{k^*}^{(f)}}, \qquad n = 0, 1, 2, \dots \;}\]
The mode $k^*$ exists only when $h_0$ and $h_f$ lie on opposite sides of the QCP $h = J$ — equivalently, the quench must traverse the equilibrium critical point. For $h_0, h_f$ in the same phase, $\cos(2\Delta\theta_k)$ has no zero on $(0, \pi)$ and $\lambda(t)$ is real-analytic for all $t$.
OBC (finite $N$) implementation
For an open chain, the Bogoliubov diagonalization yields per-site amplitudes $(\phi_n, \psi_n) \in \mathbb{R}^N$ via the Lieb–Schultz–Mattis trick
\[(A - B)(A + B) \psi_n = \Lambda_n^2 \psi_n, \qquad \phi_n = (A + B) \psi_n / \Lambda_n,\]
with the bipartite blocks $A_{ii} = 2h$, $A_{i,i \pm 1} = -J$, $B_{i,i+1} = +J$, $B_{i+1,i} = -J$. The Bogoliubov rotations $G = (\phi + \psi)/2$, $H = (\phi - \psi)/2$ map fermion operators to quasiparticles. The overlap of the H0 and Hf Bogoliubov vacua factorizes into per-mode terms whose row norms
\[\cos^2\theta_n = \sum_m |P^{(+)}_{n,m}|^2, \qquad \sin^2\theta_n = \sum_m |P^{(-)}_{n,m}|^2,\]
with
\[P^{(+)} = G^{(f)\top} G^{(0)} + H^{(f)\top} H^{(0)}, \qquad P^{(-)} = G^{(f)\top} H^{(0)} + H^{(f)\top} G^{(0)},\]
reproduce the $\cos^2\Delta\theta + \sin^2\Delta\theta e^{-2 i \Lambda t}$ factor in the diagonal-pair limit. For OBC, residual mode mixing is absorbed into per-mode $\theta_n$ consistent with unitarity ($\cos^2\theta_n + \sin^2\theta_n = 1$ enforced by row normalization to suppress round-off drift).
Public API
using QAtlas
m_0 = TFIM(J = 1.0, h = 2.0) # paramagnetic
m_f = TFIM(J = 1.0, h = 0.5) # ferromagnetic — quench crosses h = J
# Infinite (continuous-k integral, exact in N → ∞)
λ_inf = QAtlas.fetch(m_f, LoschmidtRateFunction(), Infinite();
initial = m_0, t = 1.0)
# OBC finite N (BdG diagonalisation)
λ_obc = QAtlas.fetch(m_f, LoschmidtRateFunction(), OBC(64);
initial = m_0, t = 1.0)
L_obc = QAtlas.fetch(m_f, LoschmidtEcho(:amplitude), OBC(64);
initial = m_0, t = 1.0)The pre-quench Hamiltonian is passed via the initial::TFIM keyword; initial.J must match model_f.J (only h is quenched).
Verification
test/standalone/test_tfim_loschmidt.jl exercises:
- $L(0) = 1$, $\lambda(0) = 0$.
- No-quench identity: $h_0 = h_f \implies L(t) \equiv 1, \lambda(t) \equiv 0$.
- $L(t) \in [0, 1]$, $\lambda(t) \geq 0$ across a sweep.
- DQPT cusp: $h_0 = 2 \to h_f = 0.5$ shows a slope sign-flip and a local maximum at $t_c = \pi/(2 \Lambda_{k^*}^{(f)})$, with left/right slope discontinuity dwarfing a smooth-region control point.
- OBC large-$N$ convergence to Infinite at off-cusp $t$.
- Pinned reference: $\lambda(t = 1; h_0=2, h_f=0.5) \approx 0.31693310885932685$ at
atol = 1e-8.