A switch in time sends a wave back
The episode cites a 2023 experiment in which a signal reflected in time. The lab reran the physics with two simple wave models. The effect is real and textbook: change a medium everywhere at once, and part of a wave turns round, reversed in order and shifted in frequency, while the switch pays the energy bill. Nothing arrives before it was sent.
The verdict
Real effect, different reason
Time reflection is real and matches the textbook formulas to within 0.25 %: a medium switched uniformly in time reverses a signal's order and shifts its frequency, with the switch supplying the energy. Nothing reaches a time before its cause, so it is not bending time or time travel.
Switch the wave speed everywhere at once and part of the wave travels back, with its time order reversed. The small pulse that led on the way in arrives last on the way back.
Which quantity stays continuous through the switch decides the echo. At a doubled speed the echo is +0.25 in one model and −0.5 in the other, and the simulations tell them apart.
At the switch, momentum is conserved and energy is not. The energy changes by (1 + s²)/2, and whatever drives the switch supplies or absorbs the difference.
By the rule written down before the runs, the claim would hold only for a signal that reaches a time before its cause. The only backward wave appears at the switch and after it.
The claim, as presented. The episode cites the 2023 CUNY time-reflection experiment (Moussa et al., Nature Physics 19, 863) in an episode framed around bending time and time travel.
The Why Files, “Bending Time: The Successful Time Travel Experiments using Kozyrev Mirrors”, June 2023 · the episode
The registry’s note on sources. Every claim attributed to The Why Files below comes from the episode's podcast transcript (ep. 113, podscripts.co). A research subagent summarised it (2026-09-28). The same day, before any verdict was published, the specific claims each published verdict rests on were re-checked against the transcript in a second, independent read, and both reads agree. Transcript reads go through a summariser, so treat exact wording as close paraphrase, not quotation.
Checked against the episode transcript: yes, 2026-09-28. A second, independent read of the transcript confirmed the claim as described above: the episode presents the CUNY experiment as validating Kozyrev’s view that time is not linear. The verdict is about the physics of the cited experiment, which the lab read in its preprint form.
Preregistered in PREREG.md (commit da01555) before any simulation code; one addendum (98fb77a) after run 1 and before any rerun. Self-graded: no outside model reviewed this one. The lab notebook is not public yet.
The two boundary models
Change the wave speed from c₁ to c₂ = s·c₁ at one instant, everywhere. A uniform switch keeps the wavenumber k, so the frequency is multiplied by s. What the wave does next depends on which quantity cannot jump.
utt = c(t)² ∇²u
T = (1 + 1/s)/2 R = (1 − 1/s)/2
The plain wave equation. The acceleration stays bounded across the switch, so the field and its rate of change carry straight over.
∂t(ρ ∂tu) = κ ∇²u
T = (1 + s)/2 R = (1 − s)/2
The density changes instead of the stiffness. The same step in c(t) now gives the opposite answer: at s = 2, R is +0.25 in Model A and −0.5 in Model B.
In electromagnetism the fields that stay continuous are D and B (Morgenthaler, 1958), and the D-field of a switch that changes only the permittivity follows Model A.
In a transmission line, closing the switches conserves charge, which is Model A with u = charge. Opening them keeps the voltage continuous, which is Model B with u = V. Both published formulas in the cited experiment follow from these two models.
Try the switch
Measured liveA pulse pair runs right along a line, the small one in front. Press Switch now at any moment. The page runs the same leapfrog step the lab used (dt = 0.25, dx = 1), splits the field into its right-moving part (ember) and its left-moving part (slate), and reads R and T off the field once the two have separated. The formulas sit beside each reading for comparison.
Watch the small pulse: it leads on the way in and trails on the way back. That is the reversal of time order, and it is the whole of it. The echo is a new wave born at the switch, travelling backward in space. The demo is one-dimensional and uses an instantaneous switch, like the lab’s own runs.
The experiment the episode cites
Moussa and colleagues (Nature Physics 19, 863, 2023) loaded a 50 Ω microstrip line with 30 switched capacitors that all fire within about 3 ns. Closing the switches multiplies the capacitance by 4, drops the line from 50 to 25 Ω, conserves charge and halves the frequency. Opening them conserves voltage and doubles the frequency.
An asymmetric pulse pair comes back to the input in reversed order, with its polarity flipped. The authors stress that their two switching directions obey different temporal boundary conditions, which is the Model A and Model B split above. The lab read the paper in its preprint form (main text and supplement §S5); the journal version was behind a login.
| Interface | Quantity | Paper | Lab run |
|---|---|---|---|
| Closing · Model A, s = 0.5, in volts | RV | −0.125 (Eq. 1) | −0.12502 |
| Closing · Model A, s = 0.5, in volts | TV | 0.375 (Eq. 1) | 0.37498 |
| Opening · Model B, s = 2 | R | −0.5 (Eq. 2) | −0.50059 |
Where the energy goes. For the line’s own energy, ½CV² + ½LI², the measured coefficients leave 0.625 of the incident energy at both of Moussa’s interfaces (0.6250 for Model A at s = 0.5, 0.6256 for Model B at s = 2). Each switching event takes 37.5 % of the wave energy out of the line, into the switch network. When the speed rises instead, the agent doing the switching does positive work on the wave.
The measured table
Base case: dt = 0.25, dx = 1, packets at k₀ = 2π/40. The gate was 2 % on every amplitude, for both models, at three speed ratios.
| Model | s | T measured | T formula | R measured | R formula |
|---|---|---|---|---|---|
| A | 0.5 | 1.49993 | 1.5 | −0.50007 | −0.5 |
| A | 2 | 0.75030 | 0.75 | +0.25030 | +0.25 |
| A | 3 | 0.66746 | 0.6667 | +0.33412 | +0.3333 |
| B | 0.5 | 0.74996 | 0.75 | +0.25004 | +0.25 |
| B | 2 | 1.50059 | 1.5 | −0.50059 | −0.5 |
| B | 3 | 2.00237 | 2 | −1.00237 | −1 |
Worst error 0.24 % in each model. The signs come from the fitted phase, 0 or ±π to within 1e-3 rad. Energy after ÷ before: 0.62516, 2.49936 and 4.99829 against 0.625, 2.5 and 5.
| Gate | Result | Numbers |
|---|---|---|
| G1 Model A amplitudes, 2 % | pass | worst error 0.24 % (R at s = 3) |
| G2 Model B amplitudes, 2 % | pass | worst error 0.24 % |
| G3 the wrong model’s formulas must fail | red, as intended | Model A formulas fail on Model B data, and the reverse, at every s ≠ 1 |
| G4 frequency and k | pass | ω₂/ω₁ ÷ s between 0.99995 and 1.00069; k ratios within 2.1e-4 |
| G5 no switch, no echo | pass, null | backward energy fraction 2.4e-14 (A), 3.5e-14 (B) |
| G5r break the null: a careless start | red, as intended | backward fraction 5.5e-4 against a 1e-6 threshold |
| G6 halve dt, 0.3 % | pass | largest change 0.18 % |
| G6r break it: forward Euler | red, weakly | overflows; shows only that the check can fail |
| G6r′ break it: backward Euler | red, as intended | T moves 0.0093 (A) and 0.0185 (B) between dt and dt/2; threshold 0.003 |
| G7 energy | pass | ratios 0.62516 / 2.49936 / 4.99829; work done by the switch matches the energy change |
| G8 shape and time order | pass | at a probe behind the switch, the pulse that led arrives last |
| G9 2-D refocusing, 3 % | pass | s = 2: error 0.082 %, peak on the source cell · s = 0.5: 0.015 % |
| G9n 2-D, no switch | pass, null | overlap −0.0032 and −0.0008 against a 0.05 threshold |
| G10 causality, as first built | failed | 2.5e-10 against a 1e-12 threshold |
| G10′ causality, replacement | pass · a property | backward fraction just before the switch 4.4e-14; this one cannot fail |
The failure, kept on the record. The first causality window sat 200 cells behind the large incident pulse, whose width is 30 cells. That pulse’s own Gaussian tail there is about 2.5e-10, and that tail is what the gate saw. The cause was worked out from the numbers before anything changed. The replacement, G10′, was written into the addendum and committed before it ran, and it was labelled in advance as a property that cannot count toward the verdict. So the failure does not move the verdict. It shows that a check can fail because of where the instrument sits, even when the physics is fine.
- No electromagnetic or transmission-line code was run. The EM and circuit results come from the mapping in section 01; only the two scalar models were simulated.
- The preprint’s 3 ns switching time, line dispersion and component losses were not modelled. The switch here is instantaneous and the medium lossless apart from grid dispersion.
- Only the preprint was read. Morgenthaler (1958) and the water-wave time mirror of Bacot et al. (2016) were found by search, not read.
- The episode was checked through its podcast transcript, read through a summariser, not by watching the video. Treat the wording above as close paraphrase, not quotation.
- The verdict is self-graded. No outside review was run.