Tunneling That ClicksBonus ④ / How many seconds does tunneling take?

Handling an unsettled problem carefully, while leaving it unsettled ── the conclusion that "the time cannot be defined" is probably the deeper one

How many seconds does tunneling take? Thicken the wall and the time taken does not increase (the Hartman effect).
Read naively, the effective speed exceeds light.
Relativity is of course not broken ── so what is "the time it took"?

Tools you'll need: tunneling from Episode 1, the phase of a wave, measurement from "Quantum That Clicks" The heart of this episode: it is the question that is broken

The problem Episode 1 set aside. When a particle slips through a wall, how many seconds was it inside? It looks like a plain and reasonable question. But do the calculation and something odd happens right away ── thicken the wall and the time does not increase. Read Hartman's 1962 observation straight and the effective speed rises with thickness until it exceeds light. Relativity is not broken, so something must be wrong. What is wrong is the question. There are several ways to measure "the time spent inside the wall," and they return different answers. Worse, quantum mechanics has no observable (no Hermitian operator) corresponding to "time" at all. This episode treats a question with no answer, including why it has no answer ── one episode ending in "we don't know" seems only fair.

01First, measure it naively ── the phase time

The most natural definition is to measure how much later the wave packet arrives. Compare when the transmitted peak shows up with and without the wall. This comes from differentiating the phase of the transmission amplitude (Wigner's group delay).

Phase time (group delay) $$\tau_\varphi=\hbar\,\frac{d\,\arg t(E)}{dE}$$

Compute this for a rectangular barrier and it saturates as the wall thickens. In the opaque limit

$$\tau_\varphi\ \longrightarrow\ \frac{2m}{\hbar\kappa k}\qquad(\text{independent of the thickness }a\text{!})$$

That is the Hartman effect. Double the thickness or multiply it by ten and the delay does not grow. So the "effective speed" \(a/\tau_\varphi\) can be made arbitrarily large ── and eventually exceeds light.

02Relativity is not broken ── why

The conclusion first: you cannot send information faster than light this way. There are two reasons.

Why it looks superluminal, and why that is harmless

① The emerging wave has been remade.
The wall acts as a filter that passes only the leading edge of the incoming packet, weakly, and cuts off the rest. The peak of the emerging packet is not the incoming peak that has moved; it is a reshaped fragment of the leading edge. So "the peak arrived early" does not mean "something travelled fast."

② The front of a signal always travels at exactly \(c\).
As Sommerfeld and Brillouin showed in 1914, the very leading edge of a waveform (a discontinuous onset) always propagates at \(c\), whatever the medium. Only that front can carry information. The speed of a peak is not a signal velocity.

For electrons there is also a much more concrete circumstance ── entering the apparently superluminal regime requires an extremely thick wall, where the transmission probability is of order \(10^{-1000}\). The exponential effectively seals off the superluminal region. The figure lets you check this.

03Try it ── saturation, and the seal

Figure: the phase time for a rectangular barrier (purple) and the time light takes to cross the same distance, a/c (green). The phase time saturates with thickness and eventually overtakes a/c ── formally superluminal. But the transmission probability there (red, right axis) is astronomically small
phase time τ_φ time for light to cross the same distance, a/c transmission probability (right axis)

04There is not one clock

Now the real subject. There are several ways to measure "the time spent inside the wall," and each measures a different quantity and returns a different answer.

ClockWhat it measuresFor a thick wall
phase time (Wigner)the delay of the transmitted packet's peaksaturates (Hartman)
dwell time (Smith)probability present inside the wall ÷ incident fluxsaturates, but includes reflection
Larmor clock (Baz', Rybachenko)put a weak magnetic field only inside the wall and read the time from how far the spin precessedtwo components appear; one saturates, one grows
Büttiker–Landaueroscillate the barrier and infer the time from the frequency at which it can no longer keep upgrows with thickness
weak measurement (weak values)measure weakly and average without disturbinggives a complex time

One wants to ask "which is the right one," but the question is probably posed wrongly.

The heart of this episode ── time is not an observable

In quantum mechanics, position and momentum have corresponding operators. But there is no Hermitian operator corresponding to "time" (Pauli's argument, 1933). Time is a parameter that orders states, not a thing to be measured.

So "how many seconds was it inside the wall" has no determinate meaning on its own. Meaning appears only when a specific clock is coupled to the system ── precess a spin, oscillate the barrier, track the peak of the packet. Different clocks measure different quantities.

This is exactly the measurement story from the sister series "Quantum That Clicks." "What was the particle doing inside the wall" has no answer unless you measure ── and once you measure, that measurement becomes part of the answer.

05What do the experiments say?

YearExperimentResult
1990sMicrowave and photon experiments by Nimtz, Chiao and othersObserved wave-packet peaks propagating "superluminally." Interpretation settled, however: the signal velocity does not exceed \(c\)
2008Eckle and colleagues, the "attoclock" (Science). Ionise helium with circularly polarised light and read the time from the direction the electron emergesThe tunneling delay is consistent with zero
2014Landsman and colleaguesClaimed a non-zero delay
2019Sainadh and colleagues (Nature). Repeated in atomic hydrogen, a system theory can handle accuratelyWith the Coulomb field treated properly the delay is consistent with zero. The earlier "non-zero" may have been an artefact of the analysis model
2020Ramos, Spierings, Steinberg and colleagues (Nature). Send a Bose–Einstein condensate of cold atoms through an optical barrier and measure with a Larmor clockA finite value of about 0.6 milliseconds. And they confirmed the prediction that it gets shorter at higher energy

The 2020 experiment matters. Rather than "can tunneling time be measured at all," it produced a clean answer in the form "the Larmor clock returns this value." Specify a clock and you get an answer. Do not specify one and you do not. That is where the problem currently stands.

Why the problem has lived so long Counting from MacColl in 1932, this problem has gone unsolved for over ninety years. But that is a little different from "too hard to solve."
Rather, the process of sharpening the question was itself the achievement. The plain question "how many seconds does tunneling take?" turned into "what does it mean to measure a time?" and "what can be said about paths in quantum mechanics?" ── it was the question, not the answer, that got polished. Physics has problems like that from time to time.
◇ ◇ ◇
The honest line ── almost all of this episode is "unsettled"

Established: that the phase time (group delay) for a rectangular barrier saturates with thickness (Hartman 1962); that the formal effective speed can exceed \(c\) while the signal velocity cannot (the front of a waveform always propagates at \(c\) ── Sommerfeld–Brillouin 1914); that the transmitted packet is a reshaped version of the incident packet's leading edge; that several definitions of tunneling time (phase, dwell, Larmor, Büttiker–Landauer, weak values) exist and disagree; that no Hermitian operator corresponding to time exists (Pauli); and the results of the attoclock experiments (Eckle et al. 2008, Sainadh et al. 2019) and the Larmor-clock measurement in a Bose–Einstein condensate (Ramos et al. 2020).

Unsettled, and caveats: (1) Whether there is a unique correct definition of "tunneling time" is itself unsettled. The majority position is that it has no determinate meaning without specifying a measurement, but some researchers regard a particular definition as fundamental. (2) The interpretation of attoclock experiments depends strongly on the theoretical model. The 2019 atomic-hydrogen result indicated that earlier non-zero claims may have been model artefacts, but that does not close everything. (3) The 1990s microwave "superluminal" experiments do not contradict relativity; some popular coverage was misleading. (4) The phase time in the figure is a rectangular-barrier, stationary-state, non-relativistic calculation and does not correspond to the real experimental systems (atomic Coulomb fields, optical lattices). (5) The figure's "apparently superluminal point" is a formal crossing, and the transmission probability there is displayed at the same time ── it is not an experimentally accessible region. (6) The 0.6 ms of Ramos et al. 2020 is the value that system's Larmor clock returned, not a general proposition that "an electron spends 0.6 ms inside a wall."

Exercises
  1. What is the Hartman effect?
    See the answer
    That the phase time (group delay) for a rectangular barrier saturates and stops growing as the barrier is thickened (1962). So the formal effective speed \(a/\tau_\varphi\) grows with thickness and eventually exceeds \(c\).
  2. Why is relativity still not violated?
    See the answer
    Because the peak of the transmitted packet is not the incident peak that has moved but a reshaped leading edge. Information is carried by the front of the waveform, which always travels at \(c\) (Sommerfeld–Brillouin). The speed of a peak is not a signal velocity. In addition, for electrons the apparently superluminal region has a transmission probability of order \(10^{-1000}\) ── effectively unreachable.
  3. Why is there no single answer for the "tunneling time"?
    See the answer
    Because quantum mechanics has no Hermitian operator corresponding to time (Pauli). Time is a parameter, not an observable. So "how many seconds was it inside" acquires meaning only when a specific clock is coupled to the system. Different clocks measure different quantities and give different answers.
  4. What did the 2020 Bose–Einstein condensate experiment contribute?
    See the answer
    It showed that if you specify a Larmor clock, you get a finite answer that matches the theoretical prediction (shorter at higher energy). That is: make the question concrete about "which clock" and it becomes an experimental question. Conversely, "the tunneling time" without a specified clock still has no answer.

Bonus ④ summaryNo answer comes because the question is broken

The phase time for a rectangular barrier saturates as the wall thickens (the Hartman effect, 1962), so the formal effective speed can exceed light. Relativity is nonetheless intact ── the transmitted peak is not the incident peak that moved but a reshaped leading edge, and the front of a waveform, which carries information, always travels at \(c\). Moreover, for electrons the apparently superluminal region has a transmission probability of order \(10^{-1000}\) ── the exponential seals off the superluminal region.

The deeper problem is that the definition of "the time it took" is not unique. Phase time, dwell time, Larmor time, Büttiker–Landauer time, weak values ── all return different answers. At bottom lies the fact that quantum mechanics has no time operator (Pauli). Time is a parameter, not an observable.

So the current position is this ── specify a clock and you get an answer; do not, and you do not. In 2020 Steinberg and colleagues measured about 0.6 milliseconds with cold atoms and a Larmor clock, matching the theoretical prediction. Ninety years polished the question, not the answer.

This document is Bonus ④ of the "Tunneling That Clicks" series, a reading piece for physics-loving high-schoolers and undergraduates. That the phase time for a rectangular barrier saturates with thickness (Hartman 1962), that the signal velocity does not exceed \(c\) (Sommerfeld–Brillouin 1914), that the transmitted packet is a reshaped leading edge, that several mutually disagreeing definitions of tunneling time exist, that no Hermitian time operator exists (Pauli), and the attoclock experiments (Eckle et al. 2008, Sainadh et al. 2019) and the Larmor-clock measurement in a Bose–Einstein condensate (Ramos et al. 2020) are all published results. That whether a unique correct definition of "tunneling time" exists is itself unsettled, that attoclock interpretations depend strongly on theoretical models, that the 1990s microwave "superluminal" experiments do not contradict relativity, that the figure is a rectangular-barrier stationary non-relativistic calculation not corresponding to the experiments, that the figure's "apparently superluminal point" is experimentally inaccessible on transmission-probability grounds, and that the Ramos et al. 2020 value is what that system's Larmor clock returned rather than a general proposition ── all spelled out in the body's "honest line." The phase time in the figure is obtained by numerically differentiating the phase of the transmission amplitude in the browser. ── To print, use your browser's "Print" and "Save as PDF" (in the print version the sliders and answers are frozen and hidden). Adjacent: Bonus ③, You can see a single atom because it is an exponential / Bonus ⑤, Do living things use tunneling? / Contents / sister series Quantum That Clicks and Relativity That Clicks.

Print / make a PDF: ⌘+P (Ctrl+P on Windows). On screen, move the barrier height and energy to watch the phase time saturate and cross a/c (the formally superluminal point). The transmission probability there is displayed alongside. "See the answer" opens each solution.