Continue evolution while preserving its analytic meaning
Track the time contour, the action sign, spectral damping, and reconstruction assumptions.
On this page
The question for today
What you’ll understand
- Derive the Euclidean kinetic and potential signs.
- Distinguish imaginary-time projection from thermal weighting.
- State why convergence does not by itself guarantee a unitary real-time theory.
Replacing oscillations by exponential damping often makes quantum calculations manageable. The replacement is an analytic continuation with specified boundary conditions, not permission to erase every factor of i by inspection. A regulated Hamiltonian provides a clean starting point: continue its evolution into a complex-time half-plane where the spectrum is damped. The resulting Euclidean kernel can support projection and thermal calculations, but their endpoint conditions and normalization differ. Returning to real time requires more information than a convergent numerical integral alone.
Continue the operator and action together
Start with a regulated self-adjoint Hamiltonian bounded below. Real-time evolution is exp(−iHt/ℏ). Along the continuation t=−iτ with τ positive, it becomes exp(−Hτ/ℏ). Subtracting a fixed reference energy may be convenient, but changes the overall normalization and should be declared. A finite time-sliced representation carries this continuation before a formal continuum limit is attempted; exchanging continuation, integration, and regulator removal requires an analytic justification.
For a particle with real-time Lagrangian m qdot²/2−V, the continued velocity is i times the Euclidean derivative and dt=−i dτ. Consequently the Minkowski action evaluated on that contour is i times the Euclidean action with positive kinetic term plus V. Multiplication by the original i produces minus the Euclidean action in the exponent. Rotating only the time measure while leaving the velocity unchanged would give the wrong kinetic sign.
Projection and a trace are different operations
Expand an initial state in energy eigenstates. Euclidean evolution suppresses an excitation relative to the ground state by an amplitude factor determined by the energy gap and τ. Normalizing the evolved vector therefore suppresses relative probabilities by twice that exponent. Projection requires nonzero initial overlap with the desired ground sector; a conserved symmetry can prevent that overlap. This is an imaginary-time preparation device, not unitary physical evolution preserving the original vector norm.
A thermal partition function instead traces exp(−βH), with β inverse energy and τ=βℏ. Bosonic paths are periodic around the resulting Euclidean time circle; fermionic thermal fields use antiperiodic boundary conditions. Thermal level probabilities carry exp(−βE), not the squared amplitude of an imaginary-time-evolved pure state. Confusing these constructions introduces a factor of two and can also replace a mixed thermal ensemble by a pure projected state.
Convergence is necessary but not a reconstruction theorem
A positive quadratic Euclidean kinetic operator gives convergent finite bosonic integrals, but interactions, unbounded potentials, chemical potentials, or fermion determinants may spoil a positive sampling interpretation. Gauge constraints still require gauge fixing or a quotient construction. Continuum reconstruction of a unitary real-time theory requires structural conditions such as reflection positivity and appropriate regularity, not just a small Monte Carlo error. Singularities and the vacuum iε prescription determine how time-ordered real-time boundary values are recovered.
The heat lab demonstrates spectral damping in a separate equation. Reset and set mode number to 1 and diffusivity to 1, then Step twice to time 1; predict mode amplitude e⁻¹. Reset and set mode to 2 and diffusivity to 1, then Step twice; predict e⁻⁴. Its readout also gives the spatial L² norm as √π times that amplitude. This is one periodic heat-equation mode, not a Wick-rotation engine, quantum projection algorithm, or general field-theory solver.
Work it through
Compare a projected vector with a thermal pair
A two-level system has energies 1 and 3 in units with ℏ=1. Start with equal amplitudes and evolve for Euclidean time 1, then normalize. Find its excited-level probability and compare with a thermal state at β=1.
The Euclidean amplitudes acquire factors e⁻¹ and e⁻³. Their excited-to-ground amplitude ratio is e⁻², since the gap is 2.
The normalized pure vector therefore has excited probability e⁻⁴/(1+e⁻⁴)=1/(1+e⁴)≈0.0180. Squaring amplitudes produced the factor four in this exponent.
The thermal probability is instead e⁻³/(e⁻¹+e⁻³)=1/(1+e²)≈0.1192. Both favor the ground state, but they are distinct preparations and neither result follows by treating imaginary-time evolution as unitary.
The normalized projected vector has excited probability about 0.0180; the β=1 thermal state has about 0.1192.
The living notebook
Explore the model
Interactive lab · heat
Compare analytic mode damping
Reset and set n=1,κ=1, then Step twice to t=1; predict amplitude e⁻¹. Reset and set n=2,κ=1, then Step twice; predict e⁻⁴.
- Current amplitude
- 1
- Spatial L² norm
- 1.772
- Decay rate · κn²
- 3.6 /s
The exact periodic heat-equation mode is u(x,t) = exp(−κn²t) sin(nx), on [0,2π]. Playback advances physical time. The initial sine is shown separately, and the unnormalized L² norm is √π exp(−κn²t). This analytic instrument does not run a finite-difference solver or test discretization stability. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forThe readout’s L² norm is √π times amplitude. This is an exact periodic heat mode, not analytic continuation or a quantum path-integral computation.
Pause & reason
Check your understanding
Your turn
Try it yourself
For a harmonic oscillator choose βℏω=ln4. Find the partition function with zero-point energy retained and the thermal ground-state probability.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
Sum the geometric Boltzmann series; the ratio between adjacent weights is 1/4.
Compare with the solution
The ground weight is exp[−(ln4)/2]=1/2. The sum is (1/2)/(1−1/4)=2/3. The normalized ground probability is (1/2)/(2/3)=3/4. Dropping zero-point energy would change Z to 4/3 while leaving the normalized level probabilities unchanged.
Keep the idea
Take it with you
- Wick rotation continues an operator and contour with their boundary prescription.
- Imaginary-time amplitudes and thermal probabilities have different normalization rules.
- Euclidean convergence alone does not establish a positive measure or unitary reconstruction.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.