Folding Light Back on Itself

Prologue: A Claim Worth Taking Apart

In August 2003, Michio Kaku wrote in Wired that time travel, once confined to fantasy, had become simply an engineering problem. Twenty-three years later the line circulates again, now framed as though physicists had quietly settled the question of possibility and moved on to procurement: the energy of a star, a supply of negative energy, a few paradoxes to be managed.

A reader responded with the distinction I want to make the spine of this essay. A mathematically admissible solution is not automatically a physically realizable state of nature. Mathematics offers structures; observation and physical consistency decide which of them nature uses.

I agree with that distinction. Agreeing is not enough. I want to find the exact place in the mathematics where the selection happens, write down the inequality that does the selecting, and then ask whether that inequality is a law, an assumption, or a hope.

The answer turns out to be about light. In general relativity, a time machine is a proposal to tilt the light cones until they close on themselves. The central obstruction is a statement about what bundles of light rays are allowed to do: whether they can be made to diverge without the help of matter whose energy is negative.

The circulating post makes three claims that need testing.

1. Einstein’s equations “permit” closed timelike curves, and this permission is meaningful.
2. The obstacle is the magnitude of the energy required (“the energy of an entire star”).
3. What remains beyond engineering is paradox, which is to say philosophy.

I will argue that the first claim is true but nearly empty, that the second misidentifies what is hard (the difficulty is the sign of the energy, not its size), and that the third misidentifies the cost (the paradoxes are the mildest problem; the loss of a well-posed initial value problem is the serious one). Then I will turn the same tools on the reader’s comment, and on my own premises.

Conventions. Metric signature $(-,+,+,+)$. Curvature conventions follow Wald: $(\nabla_a\nabla_b -\nabla_b\nabla_a)\,\omega_c = R_{abc}{}^{d}\,\omega_d$, $R_{ac} = R_{abc}{}^{b}$. In derivations I set $G = c = 1$ unless stated; every numerical estimate restores SI units explicitly.

Time Is the Geometry of Light Cones

In the first essay of this notebook I derived $c = 1/\sqrt{\mu_0\varepsilon_0}$ from Maxwell’s equations, and noted that $c$ is not a property of light alone but a structural constant. Special relativity completes that thought. The Lorentz invariance of Maxwell’s equations makes the quantity

the same for all inertial observers. A tangent vector $v$ is classified by the sign of $g(v,v)$: timelike if negative, null if zero, spacelike if positive. The null vectors at a point form the light cone. Causality, the statement of which events can influence which, is nothing more than the arrangement of these cones.

General relativity promotes $g_{\mu\nu}$ to a field $g_{\mu\nu}(x)$, so the cones may tilt, narrow, and widen from point to point.

2.1   Light determines nine of the ten metric components

Consider a conformal rescaling $\tilde g_{\mu\nu} = \Omega^2(x)\,g_{\mu\nu}$ with $\Omega > 0$. For any vector $v$,

and since $\Omega^2 > 0$ the sign is unchanged. The two metrics have identical light cones and identical causal structure. The converse is a theorem (Hawking, King and McCarthy 1976; Malament 1977): for spacetimes that are sufficiently well behaved causally (distinguishing), the causal structure fixes the metric up to exactly this conformal factor. A symmetric $4\times4$ metric has ten components; the causal structure fixes all but one.

So knowing which events light can connect determines nine tenths of geometry.

2.2   Light is blind to the one component it does not determine

The source-free Maxwell action in curved spacetime is

Under $g_{\mu\nu}\to\Omega^2 g_{\mu\nu}$: the determinant of a $4\times4$ matrix scales as $\Omega^8$, so $\sqrt{-g}\to\Omega^4\sqrt{-g}$; each inverse metric scales as $\Omega^{-2}$, so the product of two gives $\Omega^{-4}$; and $F_{\mu\nu} = \partial_\mu A_\nu -\partial_\nu A_\mu$ involves no metric at all. The factors cancel:

Source-free electromagnetism in four dimensions is conformally invariant. Light draws the causal skeleton of spacetime and is indifferent to the single piece of geometry that the skeleton leaves undetermined. I find this one of the quietest and deepest facts in physics. (Idealization flagged: the invariance fails with sources, with massive fields, and at the quantum level through the trace anomaly.)

2.3   What a time machine is

A closed timelike curve (CTC) is a closed curve $\gamma: S^1 \to M$ whose tangent satisfies $g(\dot\gamma,\dot\gamma) < 0$ everywhere. An observer on it accumulates positive proper time and returns to her own past.

Here is the first thing to question. By the equivalence principle, at any single point one can choose coordinates in which $g_{\mu\nu} = \eta_{\mu\nu}$ and $\partial_\lambda g_{\mu\nu} = 0$. Every local light cone looks exactly like Minkowski’s. No local experiment can detect a time machine. Chronology violation is a global property of the manifold and its cones.

That already strains the word “engineering.” Engineers act on local quantities: stress, energy density, current. The only bridge from local stress-energy to global causal structure is the pair: Einstein’s equations, plus whatever restrictions we believe real matter obeys. The whole question lives on that bridge.

Travel to the Future Is Already Engineering

Before the difficult direction, the easy one. Proper time along a worldline is

For a clock moving with speed $v$ in a weak static field, $g_{00} \approx -(1 + 2\Phi/c^2)$ and $g_{ij}\approx\delta_{ij}$ (idealization: I drop the $\Phi$-dependence of the spatial metric, which enters at higher order for slow clocks). Then

GPS as a working time machine to the future. Take a satellite at orbital radius $r = 2.656\times10^{7}$ m and Earth’s surface at $R_E = 6.371\times10^{6}$ m, with $GM_E = 3.986\times10^{14}$ m³ s⁻². Circular orbital speed: $v^2 = GM_E/r = 1.501\times10^{7}$ m² s⁻².

Velocity term: $v^2/(2c^2) = 8.34\times10^{-11}$, which over one day ($86\,400$ s) is $7.2\ \mu\text{s}$ lost.

Gravitational term: $\dfrac{GM_E}{c^2}\left(\dfrac{1}{R_E} -\dfrac{1}{r}\right) = 4.43\times10^{-3}\,\text{m}\times 1.193\times10^{-7}\,\text{m}^{-1} = 5.28\times10^{-10}$, which per day is $45.7\ \mu\text{s}$ gained.

Net: satellite clocks run fast by about $38.5\ \mu\text{s}$ per day relative to the ground (idealizations: I ignore Earth’s rotation, oblateness, and orbital eccentricity). Uncorrected, the ranging error would grow by $c \times 38.5\,\mu\text{s} \approx 11.5$ km per day.

A traveller who accelerates to Lorentz factor $\gamma$ and returns ages less than those who stayed. The cost for a $1000$ kg craft at $\gamma = 2$ is $(\gamma -1)mc^2 = 9\times10^{19}$ J, roughly a sixth of humanity’s annual primary energy use. Enormous, but finite, positive, and made of ordinary energy. For travel to the future, Kaku’s sentence is simply correct.

3.1   Why the future and the past are not symmetric

In Minkowski spacetime, along any future-directed timelike curve, $dt/d\tau = \gamma \geq 1 > 0$. The coordinate $t$ increases monotonically. Time dilation changes rates; it never reverses order.

More generally, a spacetime admits a global time function (a function whose gradient is everywhere timelike, increasing along every future-directed causal curve) if and only if it is stably causal (Hawking 1969). A global time function makes CTCs impossible: going around a closed loop, a strictly increasing function would have to return to its starting value. Travel to the past therefore requires more than extreme speed or deep gravity. It requires destroying the existence of any global time function. That is a change of kind, not of degree.

Einstein’s Equations Permit Everything

The field equations are

Now run them backwards. Choose any smooth Lorentzian metric you like, including one with CTCs. Compute its Einstein tensor. Then define

Is this source conserved? The twice-contracted Bianchi identity, which follows from $\nabla_{[a}R_{bc]d}{}^{e} = 0$ by contracting $a$ with $e$ and then $b$ with $d$, gives $\nabla^\mu G_{\mu\nu} = 0$ identically; metric compatibility gives $\nabla^\mu g_{\mu\nu} = 0$. Therefore $\nabla^\mu T_{\mu\nu} = 0$ automatically.

Every metric solves Einstein’s equations with a conserved source. Saying that general relativity “permits” CTCs, in this sense, is like saying Newton’s second law permits any trajectory because one can always define $\mathbf F := m\ddot{\mathbf x}$. True, and empty.

The physical content of general relativity lives in three places the popular claim leaves out:

1. Which $T_{\mu\nu}$ real matter can produce. Matter has its own field equations; not every tensor is the stress-energy of something.
2. Energy conditions, inequalities used as proxies for “reasonable matter.”
3. The initial value formulation. Physics asks not “does a solution exist?” but “does this spacetime develop from generic, physically reasonable initial data?”

4.1   The energy conditions, and which of them light obeys

The null energy condition (NEC) requires $T_{\mu\nu}k^\mu k^\nu \geq 0$ for every null vector $k$. For a perfect fluid, $T_{\mu\nu} = (\rho + p)u_\mu u_\nu + p\,g_{\mu\nu}$. Contracting with null $k$ kills the $g_{\mu\nu}$ term:

Since $u$ is timelike and $k$ is null and nonzero, $u\cdot k \neq 0$, so the NEC is equivalent to $\rho + p \geq 0$.

The classical scalar field. $T_{ab} = \partial_a\phi\,\partial_b\phi -\tfrac12 g_{ab}\left(\partial^c\phi\,\partial_c\phi + m^2\phi^2\right)$. Contracting with null $k$: $T_{ab}k^ak^b = (k^a\partial_a\phi)^2 \geq 0$.

The electromagnetic field. $T_{ab} = \dfrac{1}{\mu_0}\left(F_{ac}F_b{}^{c} -\tfrac14 g_{ab}F_{cd}F^{cd}\right)$. Define $l_c = F_{ac}k^a$. Then $T_{ab}k^ak^b = \dfrac{1}{\mu_0}\,l_c l^c$. By antisymmetry, $l_c k^c = F_{ac}k^ak^c = 0$, so $l$ is orthogonal to a null vector. A vector orthogonal to a nonzero null vector is either spacelike or proportional to that null vector; in both cases $l_c l^c \geq 0$. Hence

Light itself obeys the null energy condition. So does every classical field we know. This is why the NEC was long treated as nearly sacred. It is not a law, however. It is a property of classical fields, and quantum fields violate it. Hold on to that; it decides everything below.

Three Ways to Close a Light Cone

5.1   Gödel’s universe (1949)

Gödel found an exact solution sourced by rotating dust with a negative cosmological constant tuned to the density ($\Lambda = -1/(2a^2)$, $8\pi G\rho = 1/a^2$ in units $c=1$). In cylindrical-type coordinates,

The circles of constant $t, r, z$ have tangent $\partial_\phi$ with norm

This is negative when $\sinh r > 1$, that is, when $r > \ln\!\left(1+\sqrt2\right)$. Beyond that radius the $\phi$-circles are closed timelike curves. By homogeneity, every event in Gödel’s universe lies on a CTC.

Nature did not select it. Our universe has a positive cosmological constant, and the isotropy of the cosmic microwave background places tight limits on any global rotation. And notice: Gödel’s CTCs cannot be built. They are a property of an entire universe. There is no Cauchy surface on which to set initial data and evolve toward them.

5.2   Inside a rotating black hole

In Boyer-Lindquist coordinates the Kerr metric has

In the equatorial plane $\theta = \pi/2$, $\Sigma = r^2$ and

The Kerr manifold extends through the ring singularity to $r < 0$. There, multiplying $g_{\phi\phi} < 0$ by $r < 0$ reverses the inequality: $r^3 + a^2 r + 2Ma^2 > 0$. For small negative $r$ the constant $2Ma^2 > 0$ dominates, so the condition holds. The $\phi$-circles are CTCs.

To reach that region one must pass the inner horizon $r_- = M -\sqrt{M^2 -a^2}$, which is a Cauchy horizon. Radiation falling in behind the traveller is blueshifted without bound there; the Poisson-Israel mass inflation analysis indicates the inner horizon becomes a curvature singularity in realistic collapse. The honest nuance: Dafermos and Luk (published 2025) have shown the metric itself remains continuous across the perturbed Cauchy horizon, while curvature does not. So the region beyond is likely not reached as a smooth spacetime, but the question is mathematically subtle and not closed.

Here is the reader’s distinction, living inside a single solution. The Kerr exterior is a structure nature appears to select. The Kerr interior, with its CTCs, is very probably a structure nature does not.

5.3   A wormhole and a clock (Morris, Thorne and Yurtsever, 1988)

Suppose one had a short traversable wormhole with mouths A and B, initially at rest a small distance apart, with the throat’s interior synchronizing them: an event at A at external time $t$ connects through the throat to B at external time $t$.

Now move mouth B outward at speed $v$ and bring it back, taking external time $T$. Mouth B’s proper time advances by $T/\gamma$ (idealization: I neglect the brief acceleration phases). The throat is short and its internal geometry is carried along with the mouths, so the throat identifies events by the mouths’ proper times. After the trip, entering B at external time $T$ delivers you to A at external time $T/\gamma$. The jump into the past is

Let the mouths end a distance $L$ apart. Leave A at external time $t$, travel outside at nearly $c$, arrive at B at $t + L/c$, enter, and emerge from A at $t + L/c -\Delta T$. A closed timelike curve exists once

The moment $\Delta T = L/c$ is critical: the first closed null geodesic appears. Its boundary is the chronology horizon, and §8 is about what happens there.

Numbers: at $v = 0.9c$, $\gamma = 2.294$ and $1 -1/\gamma = 0.564$. A one-year round trip yields a time shift of $0.564$ yr, enough for a time machine if the mouths end less than $0.564$ light-years apart. The kinematics are elementary. The entire difficulty is the premise: a traversable wormhole.

(Other constructions exist. Tipler’s rotating cylinder needs infinite length, and Hawking’s theorem in §8 implies finite versions need exotic matter. Gott’s pair of passing cosmic strings requires a total momentum that is tachyonic, as Carroll, Farhi and Guth showed, so it cannot be assembled from ordinary sources. Alcubierre’s warp metric requires negative energy density, and a superluminal warp drive can be converted into a time machine. Every road leads to the same toll gate.)

The Throat Must Defocus Light

This section contains the one inequality that decides the question classically.

6.1   The Raychaudhuri equation for light rays

Take a congruence of null geodesics with tangent $k^a$ and affine parameter $\lambda$, so $k^a\nabla_a k^b = 0$. Define $B_{ab} = \nabla_b k_a$. Then

The first term, by the product rule, is $\nabla_b\left(k^c\nabla_c k_a\right) -\left(\nabla_b k^c\right)\left(\nabla_c k_a\right)$. The geodesic equation kills the first piece. Hence

Take the trace with $g^{ab}$ and define the expansion $\theta = B^a{}_a$. The curvature term becomes $g^{ab}R_{cba}{}^{d}k^ck_d = R_{c}{}^{a}{}_{a}{}^{d}k^ck_d = -R_{cd}k^ck^d$, using the antisymmetry of the Riemann tensor in its last pair and the definition $R_{ac} = R_{abc}{}^{b}$. Thus

Because $B_{ab}k^b = 0$ (geodesic) and $k^aB_{ab} = \tfrac12\nabla_b(k^ak_a) = 0$ (null), $B_{ab}$ effectively lives on the two-dimensional screen transverse to the rays. Decompose it there into expansion, shear and twist: $B_{ab} = \tfrac12\theta\,h_{ab} + \sigma_{ab} + \omega_{ab}$. Then $B^{ab}B_{ba} = \tfrac12\theta^2 + \sigma_{ab}\sigma^{ab} -\omega_{ab}\omega^{ab}$, and

(Subtlety flagged: the screen projection requires an auxiliary null vector; the scalar $\theta$ does not depend on that choice.)

6.2   Einstein’s equations turn curvature into energy

Take the trace of $R_{ab} -\tfrac12 R g_{ab} = 8\pi T_{ab}$ in four dimensions: $-R = 8\pi T$. Substituting back gives the trace-reversed form $R_{ab} = 8\pi\left(T_{ab} -\tfrac12 T g_{ab}\right)$. Contracting with null $k$ removes the trace term:

For a hypersurface-orthogonal congruence ($\omega_{ab} = 0$) and the NEC, every term on the right of the Raychaudhuri equation is non-positive:

Positive energy focuses light. Always. This is gravity’s attractiveness written for light rays.

6.3   Why a throat is impossible with positive energy

A wormhole throat is a minimal surface. Send a spherical bundle of light rays radially through it. Going in, the cross-sectional area shrinks ($\theta < 0$). Coming out the other side, it grows ($\theta > 0$). At the throat itself $\theta = 0$, and because $\theta$ changes from negative to positive,

By spherical symmetry the radial congruence has $\sigma_{ab} = 0$ and $\omega_{ab} = 0$. At the throat, Raychaudhuri reduces to

A traversable wormhole requires violation of the null energy condition at its throat. The throat must act as a diverging lens for light, and only negative energy can make gravity diverge light. (Hochberg and Visser generalized this beyond spherical symmetry; Friedman, Schleich and Witt proved the global version, topological censorship: in globally hyperbolic, asymptotically flat spacetimes obeying the averaged null energy condition, no causal curve can probe nontrivial topology. I cite that theorem rather than derive it.)

6.4   The concrete geometry, and the numbers

Morris and Thorne wrote the static spherically symmetric wormhole as

with throat at $r_0$ where $b(r_0) = r_0$.

Flare-out from the embedding. Take the slice $t = \text{const}$, $\theta = \pi/2$ and embed it as a surface $z(r)$ in flat cylindrical space: $\frac{dr^2}{1 -b/r} + r^2d\phi^2 = dz^2 + dr^2 + r^2 d\phi^2$. Matching,

Differentiate once more, using $\dfrac{d}{dz}\!\left(\dfrac{r}{b}\right) = \dfrac{b -rb^{\prime}}{b^2}\,\dfrac{dr}{dz}$:

A throat that flares out requires $d^2r/dz^2 > 0$ at $r_0$, which gives $b^{\prime}(r_0) < 1$.

Energy at the throat. In the static orthonormal frame, the Einstein tensor for this metric class has components (the same components that yield the Tolman-Oppenheimer-Volkoff equations for stars)

(The first follows from the general identity $G_{\hat t\hat t} = r^{-2}\,\tfrac{d}{dr}\!\left[r\left(1 -g^{rr}\right)\right]$ for such metrics, with $g^{rr} = 1 -b/r$ so that $r(1 -g^{rr}) = b$. I state the second rather than reproduce the Christoffel computation here.) With $\rho = G_{\hat t\hat t}/8\pi$ and $p_r = G_{\hat r\hat r}/8\pi$, at the throat the $\Phi^{\prime}$ term vanishes because $1 -b/r = 0$:

The embedding argument and the Raychaudhuri argument agree, as they must: the radial null vector $k = e_{\hat t} + e_{\hat r}$ gives $T_{ab}k^ak^b = \rho + p_r$.

Restoring SI units. The magnitude of the required NEC violation at a throat of radius $r_0$ is of order

With $c^4 = 8.08\times10^{33}$ and $8\pi G = 1.677\times10^{-9}$ in SI units, $c^4/(8\pi G) = 4.8\times10^{42}$ N. For $r_0 = 1$ m this is a negative pressure-plus-density of about $4.8\times10^{42}$ Pa. The central pressure of a neutron star is around $10^{34}$ Pa.

Total exotic mass. For the thin-shell construction (two flat regions glued at a sphere of radius $r_0$), the Israel junction conditions give a surface mass density $\sigma = -c^2/(2\pi G r_0)$ and total mass $4\pi r_0^2\sigma = -2r_0c^2/G$ (Visser 1989; I quote this and will derive the junction formalism in a later note). For $r_0 = 1$ m: $-2.7\times10^{27}$ kg, about minus $1.4$ Jupiter masses, or $-2.4\times10^{44}$ J.

Now compare with the post. The Sun’s rest energy is $M_\odot c^2 \approx 1.8\times10^{47}$ J. A metre-scale wormhole needs roughly a thousandth of that. The size of the energy is not the obstacle. The sign is. No quantity of ordinary energy, however astronomical, adds up to a negative number. “The energy of a star” misdescribes the problem entirely.

Where Nature Keeps Its Negative Energy

7.1   The Casimir vacuum

Classical fields obey the NEC (§4.1). Quantum fields do not. The cleanest example is the vacuum between two parallel perfect conductors separated by $d$. For the electromagnetic field, the allowed modes have $k_z = n\pi/d$. Each $n \geq 1$ carries two polarizations; $n = 0$ carries one. The zero-point energy per unit plate area is

This diverges and must be regularized. In dimensional regularization,

With $D = 2$, $s = -\tfrac12$: $\Gamma(-\tfrac32) = \tfrac43\sqrt\pi$ and $\Gamma(-\tfrac12) = -2\sqrt\pi$, so the integral equals

The $n = 0$ term has $m = 0$ and vanishes. The remaining sum becomes

using $\zeta(-3) = \tfrac{1}{120}$. The energy density in the gap is

and $F/A = -\partial_d(E/A) = -\pi^2\hbar c/(240d^4)$ is the attractive force measured in the laboratory.

Idealizations flagged: perfect conductivity at all frequencies (real metals become transparent above the plasma frequency), zero temperature, infinite plates. On the regularization: the divergent pieces are proportional to the volume and independent of $d$, so the $d$-dependent negative energy survives any physically sensible cutoff. Brown and Maclay computed the full local stress tensor between ideal plates,

so for a null vector along the plate normal, $\rho + p_z = -\dfrac{4\pi^2\hbar c}{720\,d^4} < 0$. The NEC is genuinely violated. A complete light ray crossing the gap, however, must also cross the plates, whose positive rest energy restores the integral along the full ray. Nature lends negative energy locally and collects the debt globally.

Numbers. $\hbar c = 3.16\times10^{-26}$ J·m. At $d = 10$ nm, $\rho_{\text{Cas}} = -0.01371\times3.16\times10^{-26}/10^{-32} \approx -4.3\times10^{4}$ J m⁻³. The metre-scale throat of §6.4 requires about $4.8\times10^{42}$. The shortfall is a factor of roughly $10^{38}$.

7.2   Quantum inequalities: negative energy is borrowed, briefly

Negative energy in quantum field theory is not freely available. Ford and Roman showed that for a free massless scalar field in Minkowski spacetime, the energy density sampled by an inertial observer with a Lorentzian window of width $\tau$ obeys

I do not reproduce the mode-sum proof here. Its structure is visible by dimensional analysis alone: the only density one can build from $\hbar$, $c$ and a time $\tau$ is $\hbar/(c^3\tau^4)$. It is an energy-time uncertainty relation for the vacuum: the more negative the energy, the shorter it may last.

Consequence for wormholes (my scaling estimate). Suppose the negative energy sits in a shell of thickness $\delta$ around the throat. The inequality applies when the sampling time is short compared with the local radius of curvature, so take $\tau \sim \delta/c$ (idealization: flat-space inequality applied locally in curved spacetime, which is Ford and Roman’s own justification). The throat needs $|\rho| \sim c^4/(Gr_0^2)$; the inequality allows at most $\hbar c/\delta^4$. Requiring the first not to exceed the second:

If the shell is as thick as the throat ($\delta \sim r_0$), this forces $r_0 \lesssim \ell_P = 1.6\times10^{-35}$ m. For a metre-scale throat, the negative energy must be confined to a shell of thickness $\sqrt{1.6\times10^{-35}} \approx 4\times10^{-18}$ m, some two hundred times smaller than a proton, while holding a Jupiter mass of negative energy. Ford and Roman’s detailed analysis reaches the same conclusion: macroscopic wormholes demand enormous disparities of length scale.

7.3   Averaged conditions, and the wormhole that is never a shortcut

The averaged null energy condition (ANEC) integrates $T_{ab}k^ak^b$ along a complete null geodesic and requires the integral to be non-negative. It has been proven for free fields in flat spacetime (Klinkhammer 1991) and for general interacting quantum field theories in flat spacetime (Faulkner, Leigh, Parrikar and Wang 2016; Hartman, Kundu and Tajdini 2017). In curved spacetime, ANEC can fail, but the achronal version (along geodesics no two points of which are timelike related) is conjectured to hold, and Graham and Olum argued it is what rules out time machines.

This leads to one of the most instructive recent results. Gao, Jafferis and Wall (2017) constructed a genuinely traversable wormhole using quantum negative energy, induced by coupling the two exteriors of a black hole. It is traversable. It is also not a shortcut: the signal through the wormhole arrives no sooner than causality in the full spacetime allows. Maldacena, Milekhin and Popov later built a four-dimensional version with the same character. Quantum mechanics permits wormholes you can pass through. It appears not to permit wormholes that save time. (Interpretive flag: this is my reading of how these constructions sit with achronal ANEC, not a general theorem.)

The Chronology Horizon

Return to §5.3. Suppose you somehow have a traversable wormhole and start the time-dilation manoeuvre. Before CTCs appear, a closed null geodesic appears first. The boundary of the region containing CTCs is the chronology horizon, a Cauchy horizon that Hawking called compactly generated when its null generators, followed into the past, remain inside a compact region.

8.1   Hawking’s theorem (1992)

Hawking proved that a compactly generated Cauchy horizon cannot form if the weak energy condition holds on it; the proof uses only its null part, the NEC. The structure of the argument, stated as a sketch:

1. Past-directed generators of the horizon never leave a compact set, so they must return repeatedly near themselves.
2. On each generator, Raychaudhuri with the NEC forces $\theta$ to evolve monotonically in one direction (§6.2).
3. A generator that returns to the same compact region infinitely often cannot have its cross-sectional area change monotonically forever unless $\theta = 0$, $\sigma = 0$ and $T_{ab}k^ak^b = 0$ all along it, which generic matter excludes.

Therefore the NEC must fail on the horizon. Finite time machines built in a region initially free of CTCs require exotic matter. Once again, the decisive step is Raychaudhuri for light.

8.2   The vacuum piles up

Even with exotic matter, the quantum vacuum objects. Near the chronology horizon, light can travel around a nearly closed loop. Consider a massless field whose two-point function is built by summing over images of the identification that makes the loop:

where $\sigma$ is half the squared geodesic interval (the Hadamard form) and $\Gamma$ is the isometry that carries a point once around the loop. The renormalized stress tensor comes from derivatives of the $n \neq 0$ terms at coincidence. As $x$ approaches the chronology horizon, $\sigma(x, \Gamma x) \to 0$ along the almost-closed null geodesic, and those terms diverge. Physically: vacuum fluctuations circulate the loop, and each circuit adds coherently.

Kim and Thorne argued that the throat’s defocusing weakens the divergence and that quantum gravity might cut it off at the Planck scale. Hawking argued the back-reaction still destroys the horizon. The debate is unresolved at the semiclassical level, and I will not pretend otherwise.

8.3   The theorem that ends the calculation

Kay, Radzikowski and Wald (1997) proved something sharper: on a spacetime with a compactly generated Cauchy horizon, there are points on the horizon at which no quantum state has a well-behaved (Hadamard) two-point function. The expectation value of the stress tensor cannot even be defined there.

This must be read precisely. It does not prove that nature forbids time machines. It proves that semiclassical gravity, the framework in which every calculation above was done, breaks down exactly at the moment a time machine would form. The equations stop answering at the door. Whatever decides the question lies in quantum gravity. Hawking’s chronology protection conjecture, that the laws of physics prevent the appearance of CTCs, remains a conjecture. (He liked to add, half in jest, that the absence of tourists from the future is observational support.)

The Real Price: Losing the Cauchy Problem

The post lists “paradoxes of causality” as the remaining challenge. I think this places the problem in the wrong layer.

9.1   Global hyperbolicity

A spacetime is globally hyperbolic if it has a Cauchy surface: a slice that every inextendible causal curve crosses exactly once. Geroch proved that global hyperbolicity is equivalent to the existence of such a surface, and it is precisely in globally hyperbolic spacetimes that Choquet-Bruhat’s theorem guarantees Einstein’s equations have a well-posed initial value problem: data on the slice determine the evolution uniquely and stably.

A spacetime with CTCs is not globally hyperbolic. A closed timelike curve would cross any candidate Cauchy surface either never or repeatedly. The logic of physics as prediction, data now determining what comes next, is suspended.

9.2   Paradox is not the problem; excess is

Polchinski proposed the sharpest paradox: a billiard ball enters a wormhole time machine at an angle such that it emerges in the past and knocks its earlier self off course, so that it never enters. Echeverria, Klinkhammer and Thorne analyzed it and found something unexpected. For the dangerous initial conditions, there exist self-consistent trajectories (the ball receives a glancing blow that sends it into the mouth on exactly the path that delivers that glancing blow). Worse, for the same initial data there are infinitely many self-consistent solutions.

So the classical difficulty is not that time machines produce contradictions. It is that they produce too many consistent histories, and the laws no longer choose among them. Friedman and collaborators showed that for a broad class of such problems solutions exist but uniqueness fails. Predictivity is what is lost.

Novikov’s self-consistency principle, that only globally consistent histories occur, is often offered as the answer. I want to question it. Stated as a law it is nearly a tautology (whatever happens, happened). Stated as a selection rule it is silent on which of infinitely many consistent histories occurs. It names the problem rather than solving it.

9.3   The quantum version

Deutsch (1991) proposed that a quantum system on a CTC must satisfy a fixed-point condition. If the chronology-respecting system A is in state $\rho_A$ and interacts with the CTC system via unitary $U$, the CTC state must satisfy

A solution always exists. The set of density matrices on a finite-dimensional Hilbert space $\mathbb{C}^d$ is a compact convex subset of $\mathbb{R}^{d^2 -1}$. The map on the right is continuous in $\rho_{\text{CTC}}$ and maps that set into itself. By Brouwer’s fixed-point theorem, it has a fixed point.

The grandfather paradox as a qubit. Let the CTC carry one bit, and let the interaction flip it: whatever arrives from the future prevents itself, so $U$ acts as Pauli $X$ on the CTC qubit. The condition becomes $\rho = X\rho X$. Write $\rho = \tfrac12\left(I + r_xX + r_yY + r_zZ\right)$. Conjugation by $X$ sends $(r_x, r_y, r_z)\to(r_x, -r_y, -r_z)$, so the fixed points are $r_y = r_z = 0$, including the maximally mixed state $I/2$ that Deutsch’s maximum-entropy rule selects. The paradox dissolves into a mixture: each history occurs with probability one half.

The price. The output state of system A depends on $\rho_{\text{CTC}}$, which depends on $\rho_A$. The evolution of A is therefore nonlinear in its initial state. Quantum mechanics is linear. Deutsch CTCs would allow distinguishing non-orthogonal states and, as Aaronson and Watrous proved, would make classical and quantum computers with CTCs equivalent to PSPACE. Lloyd and collaborators proposed an alternative model based on post-selection, which gives different predictions. There is no unique quantum theory of time travel. That alone tells us the question is not engineering: we do not yet know what the laws would be.

Questioning the Post, the Comment, and Myself

10.1   The post

• “Officially crossed over into practical science.” There is no such official status. The source is a 2003 magazine essay by one physicist.
• “Rather than debating whether the laws of physics forbid such journeys…” This is the reverse of the truth. From Morris, Thorne and Yurtsever in 1988 through Hawking’s theorem, the quantum inequalities, and Kay, Radzikowski and Wald, the literature is the debate about whether the laws forbid it.
• “The energy equivalent of an entire star.” The magnitude is not the obstacle; the sign is (§6.4).
• “An engineering problem.” An engineering problem presupposes known physical principles and admissible materials. Here we do not know whether the required state of matter exists at all, and the governing theory at the decisive moment (quantum gravity at the chronology horizon) does not yet exist.

To be fair to Kaku: his underlying point is that no known law flatly and provably forbids time travel, and that is true. No theorem closes the door in full quantum gravity. But “not proven impossible” and “only engineering” are separated by the entire content of §§6 to 9.

10.2   The comment

The reader’s central distinction is correct and important. I would sharpen one part of it. The comment groups black holes with singularities, wormholes, and other extreme solutions. That grouping hides the most instructive fact in this whole subject.

Penrose’s theorem (1965), from the same equation. Suppose a trapped surface forms, so that outgoing light rays have $\theta_0 < 0$. With the NEC, §6.2 gives $d\theta/d\lambda \leq -\tfrac12\theta^2$. Then

Since $1/\theta_0 < 0$, the right side reaches zero at $\lambda = 2/|\theta_0|$, so $\theta \to -\infty$ within that affine distance: the rays reach a caustic. Combined with a global argument (a non-compact Cauchy surface), Penrose showed that spacetime must be null-geodesically incomplete.

Set this beside §6.3. The single inequality $T_{ab}k^ak^b \geq 0$, fed through the Raychaudhuri equation, makes black holes inevitable and makes traversable wormholes impossible. Black holes are forced by ordinary matter; wormhole time machines are forbidden by it. They sit on opposite sides of the same sign.

And the observational record matches. The gravitational-wave signal GW150914 and the many mergers since have ringdowns consistent with Kerr exteriors at the precision tested; the Event Horizon Telescope has imaged shadows around M87* and Sagittarius A* consistent with the Kerr prediction. Black hole exteriors are, on present evidence, structures nature has selected. Where the reader’s caution does bite hard is in the interiors: the singularity (Penrose predicts incompleteness, not infinite curvature, and quantum gravity is expected to change the story) and the Kerr inner horizon with its CTCs (§5.2). The black hole is not an example of an unselected solution. It is the clearest example of one solution containing both.

A framework I would propose. “Observation decides” is true but incomplete, because we have not observed the absence of the Kerr inner region; we infer it. I would name three filters that a mathematical solution must pass to count as a physical state of nature:

1. Dynamical accessibility. Does it develop from generic, physically reasonable initial data on a Cauchy surface? (Black hole exteriors: yes. Gödel: there is no Cauchy surface to start from.)
2. Stability and quantum consistency. Does it survive perturbation and the quantum back-reaction of its own fields? (Kerr inner horizon: mass inflation. Chronology horizons: vacuum pile-up and the Kay-Radzikowski-Wald breakdown.)
3. Observation. Where the first two leave room, does nature show it to us? (Kerr exteriors: ringdowns and shadows. Traversable wormholes: nothing.)

Time machines fail the first filter classically and run into the second quantum mechanically. They never reach the third.

10.3   Myself

Now the premises of this essay.

Is the NEC a law? No. §7.1 exhibits its violation in a laboratory measurement. My argument therefore cannot rest on the NEC, and it does not: it rests on the quantum inequalities and on averaged conditions.

Is ANEC a law? In flat spacetime, for quantum field theory, it is now a theorem. In curved spacetime it can fail, and the achronal version that would forbid time machines is a conjecture. The strongest selecting principle available is unproven in exactly the regime where it is needed.

Is chronology protection a law? It is a conjecture supported by semiclassical calculations that themselves break down at the decisive moment.

Does full quantum gravity allow topology change, spacetime foam, or causal structure that is not a smooth Lorentzian manifold at all? Nobody knows.

So I will not claim that time travel to the past is impossible. I will claim something narrower and, I think, defensible. Every route found so far requires negative energy in amounts, durations, and geometric configurations that quantum field theory either forbids or allows only in forms that are never shortcuts, and every semiclassical attempt to form a chronology horizon destroys the framework used to describe it. That is not a proof. It is a map of the wall, drawn from many directions, and every direction finds the wall in the same place: the vacuum.

That is where this essay joins the rest of this notebook. The question “can we build a time machine?” turns out to be the question “can the quantum vacuum be sculpted to lend negative energy globally and durably?” It is the same vacuum whose energy density disagrees with observation by some 120 orders of magnitude, and the same vacuum whose Casimir stresses I examine elsewhere in this series. The time machine is not an engineering problem. It is a vacuum problem, and the vacuum is the least understood object in physics.

Coda: Light Does Not Return to Its Own Past

In §2 I showed that light determines the causal skeleton of spacetime and is blind to everything else. Every time machine is therefore a proposal about light: bend its cones until a ray returns to where it began, and keep bending until something slower than light can follow.

What stands in the way is not a lack of fuel. It is that gravity, fed by any energy that is positive, always focuses light. To make light diverge, to open a throat, to let a ray come home before it left, one needs energy below zero, and the quantum vacuum lends that only in thin slivers, briefly, and always takes it back.

Einstein’s equations write down every geometry one can imagine. Nature reads them as a list of possibilities and edits with a single sign.