To the editors:

Daniel Jassby has two main conclusions in his recent essay concerning the production of electric power by fusion energy: “It is likely unachievable anytime in the next half a century,” and “ICF [inertial confinement fusion] appears to be a far more likely candidate than MCF [magnetic confinement fusion] as the basis for a power plant.” My purpose here is to explain why the first might be too pessimistic because of recent engineering progress and why the second is certainly incorrect because it does not take seriously inertial fusion energy’s engineering challenges.

Fusion reactions power the stars. The conditions needed for fusion to occur are known. Producing fusion reactions artificially is also fairly easy for a high-tech society: such reactions are the basis of the hydrogen bomb. But achieving controlled fusion energy for power production is vastly more difficult. It is certainly far more difficult than the controlled fission reactors that for 40 years have provided about 20% of the electricity on the US grid. Consider: it took only about four years between the scientific discovery of nuclear fission—the breaking up of uranium nuclei by neutron bombardment—and the operation of the first self-sustaining controlled fission reactor. By contrast, seventy years and the prolonged efforts of thousands of scientists have yet to achieve self-sustaining controlled fusion. It remains, nevertheless, a grand challenge in science and engineering that continues to inspire the imaginations and efforts of succeeding generations of students and researchers, not least because it holds out the possibility of an important, clean, and sustainable energy source.

Magnetic confinement fusion, or MCF, is more usually referred to as magnetic fusion energy (MFE). It aims to contain an incredibly hot gaseous plasma, roughly 100 million degrees Celsius, by the force of a carefully crafted magnetic field configuration. Jassby’s first conclusion appears to be based on his assessment that the approach of magnetic confinement is suffering from “stagnation and retreat.” Fusion plasma confinement must be good enough that plasma heat and particles escape more slowly than they react. Otherwise, like a pile of damp sticks to which the flame of a match is held, the reaction will fizzle out when the match—in this case, externally supplied plasma heating—is removed, because the reaction heat cannot overcome losses.

The tokamak magnetic configuration has, since the 1960s, been the most successful at minimizing plasma loss and maximizing confinement. Over the succeeding years, scientific understanding of the loss processes has developed enormously. It is now known within a factor of two or so what it will take to achieve ignition or the burning plasma state quite close to it. The largest tokamaks to have operated—the Tokamak Fusion Test Reactor and JT-60U, which have both closed down, and the Joint European Torus (JET), which is still operating—fall short of the required magnetic configuration parameters. That requirement, simplified drastically and with apologies to my professional colleagues who know the enormous complexities I am glossing over, is that the product of magnetic field strength, B, and the major radius of the plasma torus, R, must exceed roughly 20 tesla-meters.¹ JET has a BR product of approximately 10. The Tokamak Fusion Test Reactor and JT-60 had similar values. None are sufficient to enter the burning plasma regime. And since no experiment with greater BR has been built since, the maximum confinement performance achievements of MFE have, I accept, arguably been stagnant since the 1990s, even if scientific understanding and many other aspects of performance have not. JET’s recent deuterium-tritium (D-T) campaign achieved comparable plasma performance to its 1997 experiments, but did so with vital changes to plasma-facing materials and components, as necessary for the success of future D-T experiments and reactors.² That achievement deserves to be regarded as major progress.

The International Thermonuclear Experimental Reactor (ITER), a giant tokamak under construction in France by a collaboration of 35 nations, is designed to achieve and study burning plasmas. It will have approximately B = 5 tesla and R = 6 meters, for a BR of 30. But ITER’s great size and complex organization have made it expensive and painfully slow to construct. It will not operate until 2026 at the earliest and, because of complexities surrounding radioactivity, will not use the tritium fuel needed for burning plasma until probably the mid-2030s. On that timescale, Jassby’s first conclusion would be inescapable. For nearly 30 years, I have been answering the “when will we have fusion” question with “not on the grid in my lifetime.” If ITER is the only option, my prediction is safe. But don’t misunderstand my view. ITER is the product of a research program I have been involved with most of my career. The national experimental facility at MIT I led for many years helped with the specification and design of ITER. I expect ITER will eventually be an important scientific facility and demonstrate a controlled fusion burn yielding hundreds of megawatts of fusion power, lasting hundreds of seconds at a time.

It is not because of ITER that I think Jassby’s first conclusion might be too pessimistic. It is that, in the past year, a team from MIT and Commonwealth Fusion Systems has demonstrated that, by using high-temperature superconductors, much higher magnetic field strength in the tokamak configuration can be generated without significant dissipation.³ This increase makes the proposed SPARC experiment feasible, which is a design with roughly B = 10 and R = 2.2, so BR = 22.⁴ On paper, SPARC can demonstrate burning plasma performance in a machine whose volume and mass, and hence cost, are roughly 25 times smaller than those of ITER. SPARC can also be built much faster. Already its estimated construction cost of about US$2 billion has been raised from venture capital, and its construction is proceeding. It is aimed—optimistically, in my opinion—to operate in 2025 and to achieve a burning plasma years earlier than ITER.

Inertial confinement fusion is how hydrogen bombs work. It relies on reactions at densities roughly a trillion times higher than magnetic fusion and taking place so quickly that most of the fuel reacts before the high-pressure assembly explodes apart. To produce useful energy, as opposed to devastation, the idea of inertial fusion energy (IFE) is that each explosion’s size should be scaled down to something that is manageable and repeated once per second or faster to provide a source of useful average power. That might sound a bit crazy, but currently almost all the power for ground transport works on the principle of repetitive explosions: the internal combustion engine. It is debatable how small the energy yield of each inertial fusion explosion must be to make it manageable, but optimistic estimates usually take it to be a billion joules. That is the same energy yield as approximately 20 kilograms of TNT—a big explosion. A smaller upper limit is more plausible. IFE also must demonstrate a method of causing fusion micro-explosions without the use of a fission bomb trigger. Delivering the required extreme triggering power is most easily done with high-power lasers—hence the colloquial name “laser fusion.”

ICF has been supported for decades, at funding levels rather higher than MFE, for the purposes of “stockpile stewardship”: the maintenance of nuclear weapons and related expertise in US national laboratories. IFE research is a sideline that benefits scientifically from this effort. The recent achievement of a fusion yield of 1.3 megajoules in a single pulse on the big laser-based National Ignition Facility (NIF) at the Lawrence Livermore National Laboratory is a big step forward in performance compared with where NIF has struggled since coming into operation in 2009.⁵ The scientists and engineers who worked to achieve it are to be congratulated. Their efforts have finally demonstrated ignition, or close to it, and overcome many complicated problems that prevented the ignition performance that was confidently predicted for the first National Ignition Campaign, which ended unsuccessfully in 2012.

A crucial aspect of the physics challenges Jassby fails to convey is that, to produce net energy, IFE must have confinement performance considerably greater than what is needed for ignition. It must be sufficient to ensure that almost all the fuel is burned in each pulse. The reasons, as well as the overall character of pulsed fusion, can be deduced from elementary undergraduate physics estimates and are not controversial.⁶ Fusion confinement performance is technically best evaluated by the Lawson product of density and confinement time. In ICF, it is conventionally expressed as the product of mass-density ρ and assembly radius R, because the confinement time can be considered to be the time taken to travel a distance R at a known speed. The ρR required for D-T ignition is approximately 0.3 grams per square centimeter, but the value required for high gain, to burn at least 50% of the fuel, is approximately ten times larger than ignition: 3 grams per square centimeter. By contrast, in MFE, all that is required is ignition or just a fraction less. Accepting that NIF has last August—maybe just—achieved ignition, it is still a factor of 10 short, in Lawson parameter ρR, of what is needed to generate net energy. JET’s 1997 Lawson parameter was only a factor of 5 short of what is needed in MFE. It was arguably closer, even then, than IFE is today.

It is, though, mostly by a tendentious assessment of the engineering practicalities that Jassby arrives at his second conclusion, that IFE is more likely to succeed than MFE. He emphasizes the substantial tritium inventory of JET, regarding its safe management as a liability rather than a success and saying that tokamaks need “1,000 times as much tritium per pulse as ICF experiments.” This is a spurious comparison since the pulse length of JET is approximately a trillion times longer than that of ICF. The amount of tritium consumed in producing one joule of fusion energy is exactly the same for MFE as for IFE. He argues that the modest amount of tritium in each small ICF capsule “makes it practical to use tritium on every shot,” but neglects to note that NIF is “limited to 20–30 ignition shots each year,”⁷ while JET can run 30 shots per day.

Jassby rightly points to one of the major technical challenges of MFE as being plasma interaction that possibly damages or erodes the chamber walls, but dismisses the even greater materials interaction challenges in IFE. The “one-meter-thick flowing liquid metal sphere or cylinder” he invokes as IFE’s solution is an idea no more practical for IFE than it is for MFE.

Returning to tritium problems, he supposes, in contradiction to many studies, that tritium self-sufficiency is impossible and that therefore “reactors must be fueled by deuterium alone.” No responsible fusion expert now thinks D-D fusion energy is possible for tokamaks. But for Jassby to dismiss the comparable conclusion for IFE, which requires pellet masses probably at least 20 times larger than D-T, by saying “the driver energy must be as large as 100MJ, which is feasible with several of the drivers listed previously,” is itself to enter a fantasy world.

In reality, terrestrial controlled fusion energy production is extremely difficult and might not be practical. Humankind has, over roughly seventy years of research, reached the threshold of the burning plasma regime where self-heating of the reacting plasma predominates over applied heating. This achievement by both inertial and magnetic confinement is remarkable but is still a substantial distance from scientifically demonstrating a self-sustained controlled fusion reaction sufficient for energy production. That demonstration might be achieved in MFE by ITER or SPARC. It is very unlikely to be demonstrated on NIF because, in ICF, ignition is not sufficient for IFE. Even if such demonstrations are achieved, harsh engineering challenges will need to be overcome before a fusion reactor can produce electricity. Those challenges are substantially greater for IFE, in my opinion, than they are for MFE. Even so, MFE has already begun to address engineering in ways that IFE has not, because the MFE burning plasma state is of the same scale as for a reactor, while for IFE large scale-ups or even total transformations are required relative to burning plasma experiments: in confinement, repetition rate, explosion energy, debris management, driver efficiency, and so on.

Over the years, many proposed approaches to fusion energy have been shown to be impossible because of the laws of physics. Current experiments are not at a place where anyone can or should choose final winners and losers among those that remain. But research resources should be directed into those that, on the basis of measured scientific and engineering judgment, offer the best practical prospects for success. Jassby gives a mostly measured scientific assessment, but falls short when it comes to engineering. In the end, it may well be engineering, not science, that determines whether or not fusion energy is feasible.

Ian Hutchinson

Daniel Jassby replies:

In response to Ian Hutchinson’s letter, I will continue to use the abbreviations ICF and MCF where appropriate, because my essay is concerned with the scientific feasibility of thermonuclear plasmas, not with the production of electrical energy, for which the terms IFE and MFE are more pertinent.

Hutchinson’s view of the status of MCF is overly sanguine, partly because he has ignored my essay’s discussion of beam-thermal versus thermonuclear fusion in JET. His letter also underplays the debilitating consequences of various tritium-related issues in MCF devices, which are discussed herein.

Hutchinson states that “JET’s 1997 Lawson parameter [the product of density and confinement time] was only a factor of 5 short of what is needed in MFE,” while the highest equivalent value achieved for NIF is a factor of 10 short. Actually there are many fusion experimental devices of all types that have huge Lawson parameters, but for one reason or another cannot demonstrate high energy gain Q and fusion neutron output. The latter are the benchmarks that matter in practice. In any case, the more commonly used parameter is the fusion triple product, defined as the product of the Lawson parameter and ion temperature. Compilations of experimental data show that this metric was an order of magnitude higher for NIF than for JET even before the August 2021 supershot.⁸

Hutchinson errs in stating that “both inertial and magnetic confinement” have “reached the threshold of the burning plasma regime.” While numerous shots on NIF have actually entered the burning plasma regime,⁹ the JET device in 2021 was not close to breakeven, Q = 1, even when including beam-thermal reactions. Only thermonuclear reactions can be scaled to the high Q-values needed for electrical energy production, and the thermonuclear Q of JET did not exceed 0.2 on a quasi-steady basis in the 2021 campaign, as discussed in my essay. When only thermonuclear reactions are considered, JET’s performance was a factor of 5 from breakeven and a factor of 25 below the Q = 5 that denotes a burning plasma. The 1997 shots were somewhat better—hence the characterization “stagnation and retreat.”

Hutchinson is of course correct that to obtain a practically useful level of Q for ICF (Q ~ 100), the areal density ρR of the compressed fuel capsule must be increased by a large factor and the tritium burnup must be of order fifty percent. But the required large tritium burnup is actually a great advantage of ICF, because it permits a small amount of the unburned tritium to be lost without consequence. In contrast, any small loss of the unburned tritium in an MCF reactor with its characteristically minute tritium burn-up is devastating to fuel replenishment, which must be generated by absorbing neutrons in the lithium blanket that surrounds the plasma.

Here is an example. Define f_b as the fractional tritium burnup per pass, that is, the fraction of injected tritium that is consumed in fusion reactions before it leaves the reaction vessel. Suppose f_b is 1%, typical of future large tokamaks. Then 99% of the injected tritium must be recovered by pumping the plasma exhaust and by scavenging, scouring and heating all the plasma-facing surfaces including appendages. If just 1% of this 99% cannot be recovered because it is embedded in a surface or has diffused into the reactor structure, then the global tritium breeding ratio must be 2, which is impossible. But if f_b is 50% as in ICF, and 1% of the remainder cannot be recovered, the global tritium breeding ratio must be only 1.02. In both cases, loss of neutrons through reactor vessel penetrations is neglected, but is likely on the order of 10% and important.

Hutchinson helpfully points out that “the amount of tritium consumed in producing one joule of fusion energy is exactly the same for MFE as for IFE,” but in the tokamak example with f_b of 1%, a given amount of tritium must be scavenged, pumped, segregated, and re-injected by an array of complicated processing systems 100 times before complete burnup. With a 1% tritium loss on each grand tour, the MCF reactor would in effect consume twice as much tritium as the ICF reactor.

Hutchinson is correct that I claim tritium self-sufficiency is impossible, in contradiction to many studies. Those theoretical studies totally ignored unrecovered tritium, a major problem in both the Tokamak Fusion Test Reactor (TFTR) and JET.¹⁰ In recent years that problem has been recognized but not addressed. Mohamed Abdou and his collaborators, who are probably the leading designers of reactor breeding blankets, emphasize that, in addition to a host of difficult constraints to be satisfied, tritium replenishment is not possible unless f_b is at least 5%—a very tall order for tokamaks.¹¹

In the D-T campaigns in TFTR and JET, f_b averaged an astoundingly low 0.01%, even after exempting those shots used for “wall conditioning,” or for pure-tritium plasmas, or aborted by disruptions. In the much-publicized 59MJ shot of 2021,¹² the JET project boasted that it burned up only 0.1mg of tritium, neglecting to mention that almost 1 gram was injected during the pulse and far more than 0.1mg was no doubt irretrievably lost in the reactor structure.

Tokamak plasmas with Q ≫ 1 will have values of f_b that are many times greater than JET’s value, but will struggle to reach 5% and likely cannot.

It is ironic that ICF systems with their requirement, as Hutchinson points out, that f_b be of order 50% can easily meet the 5% criterion, but in the long run may be able to avoid external tritium breeding altogether, as discussed in my essay.

Hutchinson notes that JET can fire dozens of times more shots than NIF, but the frequency of high-power D-T shots in JET is actually not so different from the NIF shot rate. JET can run many shots per day using non-tritium gases when it is discharge cleaning plasma-facing surfaces, a periodic exercise that is forced on all tokamaks. In each of its D-T campaigns, JET fired approximately 200 high-power D-T shots over a four-month period, averaging about two shots per day.¹³ An occasional run day might have five or six full D-T shots. NIF can fire one shot per day at its highest energy output, but is restricted administratively to about 90 ICF-related shots per year, including 20 to 30 ignition shots, because of its many other mandatory applications, such as stockpile stewardship and astrophysical simulations.¹⁴ Those 90 shots can all use tritium.

While ICF experimental devices regularly employ tritium, the next use of tritium in MCF will occur only in the 2030s, in the ITER and SPARC tokamaks.

Hutchinson pins all his hopes on the SPARC tokamak to speed the early realization of fusion reactors. A project of Commonwealth Fusion Systems and the Massachusetts Institute of Technology, the SPARC device will have B = 12T and R = 1.85m,¹⁵ not the 10T and 2.2m that Hutchinson quotes. The distinction is important, because the electromagnetic stresses on the magnets that must be counteracted and the synchrotron radiation losses from the plasma are both 45% larger at 12T than at 10T.

SPARC will have no beam injection. Instead it will rely solely on ion cyclotron radiofrequency (ICRF) heating of minority ion helium-3 embedded in the D-T mixture. The closest JET experience that has been published was its 1997 campaign with ICRF heating of minority-ion populations in D-T plasmas.¹⁶ The highest quasi-steady Q obtained was 0.22. This was achieved using a technique that energized a 10% deuteron population to many tens of kilovolts up to 120keV in a plasma that was 90% thermalized tritium, so that the fusion reactions were predominantly of the beam-thermal kind. In plasmas with greater deuterium content, ICRF heating produced a Q of about 0.1 with mainly thermonuclear reactions. As it happens, Q = 0.1 to 0.22 is similar to the range of thermonuclear Q demonstrated in JET’s 2021 campaign with beam injection.

To produce a burning plasma—that is, Q = 5—the SPARC project must achieve a Q some 20 times larger than JET’s best thermonuclear values of 0.2 to 0.25. All of the improvement, as well as compensation for smaller plasma major and minor radii that are respectively 2/3 and 1/2 of JET’s, has to come from maximizing plasma current and density to take full advantage of the three times higher B. Assuming that engineering solutions to counteract huge magnet stresses allow practical operation at 12T, Q must increase as the third to fourth power of B. This dependence is fortunately consistent with empirical scaling laws for energy confinement that predict maximum achievable Q varies approximately as R² × B⁴.¹⁷

A far different result is found when scaling against the performance of MIT’s Alcator C-Mod tokamak, the high-field precursor to SPARC that operated with R = 0.67m and B = 8T in deuterium plasmas. The maximum Q in D-only operation that was attained using ICRF with minority-ion heating—essentially the same heating technique planned for SPARC—was 6 × 10^–5, or 300W of fusion with 5MW of heating.¹⁸ Those numbers extrapolate to an equivalent Q = 0.015 in D-T operation. Now the improvement in Q in extrapolating to SPARC comes predominantly from increase in size, but the R² × B⁴ scaling predicts a Q of only 0.6 in SPARC. Empirical scalings can evolve. If Q would increase as R³ × B⁴ rather than R² × B⁴, extrapolations from C-Mod and JET to SPARC give comparable values of Q = 2 and 3, respectively.

Thus SPARC is a little too small to achieve a burning plasma, even with ideal operating conditions. It can work toward obtaining those conditions only in non-tritium plasmas.

A highly limited number of D-T shots will constrain progress on the SPARC device. The project has committed to an on-site inventory of only ten grams of tritium, a concession that Commonwealth Fusion Systems had to make to win state regulatory approval for operation in a highly developed location. It will have a lifetime tritium throughput on the order of 100g, similar to JET’s. And like JET it will have only about two hundred D-T shots with 5- to 10-second pulse length, which will permit little opportunity for plasma optimization.

In line with the TFTR and JET experience, SPARC will have to depend on hydrogen, helium, and deuterium plasmas for optimizing plasma conditions and developing techniques to mitigate adverse plasma-surface interactions, such as overcoming influx of tungsten impurities and minimizing plasma disruptions. It may take many years to develop the ideal conditions in deuterium that the experimentalists hope will lead to the desired high-Q performance in D-T. Even then, the outcome of switching to D-T operation is always uncertain because of the vagaries in tritium fueling, changes in plasma energy confinement with isotope mix, and effects of fusion-produced alphas that will be encountered only with D-T plasmas.

If SPARC is very successful, it will generate 0.5 to 0.8GJ of fusion energy per pulse, consuming just 1mg of tritium. But it will have to inject approximately 1g of tritium per pulse, achieving a fractional burnup of merely 0.1%. If f_b = 5% is the threshold for ensuring fueling viability, a successful SPARC will fall short by a factor of 50, while the NIF supershot has achieved f_b = 1.5%, within a factor of 4 of this criterion.

Hutchinson aptly observes that, in the end, engineering will determine whether fusion energy is feasible. That is why I closed my essay with the statement that “[t]he technological hurdles for implementing an ICF-based power system are so numerous and formidable that many decades will be required to resolve them—if they can indeed be overcome.” As for the MCF enterprise, irremediable tritium travails and chronic inability to isolate the plasma from grievous interaction with its immediate environment may decisively derail progress before reactor engineering challenges need be surmounted.