IARPA Withdraws Funding From Major NIST Quantum Computing Groups

Do you think this is a way to twist their arms to go "underground" and work on classified stuff?

Dave:

Your question amounts (in part) to the broader question "What's sufficiently good about QIT/QIS that IARPA and similar agencies should invest in it?"

In this regard, three thought-provoking sources for further reading are:

(1) Business Week's cover story this week The Failed Promise of Innovation in the US (June 3, 2009).

(2) The in-depth web site Small Wars Journal (the US is at war, remember?).

(3) David Petraeus' article in the Sep-Oct 2008 Military Review, Commander's Counterinsurgency Guidance.

Let's ask, what does QIS/QIT have to do with the subject matter of these three documents? Because ... hmmmm ... if the answer is "not much", then perhaps agencies like IARPA should realign their research programs.

Now, according to Gen. Petraeus' Guidance: "In the absence of guidance or orders, determine what they should be and execute aggressively."

So without waiting to be told, we should ask (and answer) for ourselves the question what should QIS/QIT researchers be working on? Because as the directive says, we shouldn't wait to be told ... we should just start working on it. :)

In this regard, I've started posting my QIS/QIT technical questions on Wilmott ... it's fun forum ... and the mathematical maturity of Wilmott readers is pretty darn high. A Google search for "Commutative noise in finance versus biomicroscopy" will find the thread.

The Wilmott readerships understand quite well the key role of math/science/engineering in reducing the risk, speeding the pace, and increasing the returns of enterprise ... three themes that the documents that began this thread also focus upon.

You'll notice that I haven't answered the question "What should QIS/QIT researchers be working on?" That's not because I don't have my own answers, but because I'm very interested in other folks' answers.

Cowabunga, as ever ... John Sidles

The paucity of comments on this thread is surprising!

A comparative analysis of the (above-named) 2009 Business Week cover story with this 1957 Time Magazine cover story The New Age

http://www.time.com/time/magazine/article/0,9171,821134-1,00.html

speaks volumes about the present-day enterprise of math, science, engineering, and technology.

There is a new person in charge at IARPA and she does not think that they should be giving money to federally funded research labs. A big QKD effort at LANL recently lost its IARPA funding as well.

While surprising when it affects big names like these, this sort of thing is always happening at some rate, for a couple of reasons. First, when leadership at a DOD organization changes, funding priorities almost always change to some extent - these individuals have lots of power to steer resources, and it's not surprising that there are differences of opinion. Second, funding agencies really do like to have "ownership" of topics. They want to be able to make the case to Congress that their support for an area was critical to a major advance, and that's easier to do when there is no sharing of resources with other agencies. Third, it's possible that IARPA has decided that their money could be better spent on other things because quantum computing, while potentially cool for NSA codebreaking, won't help them prevent some fanatics with a truckload of semtex from blowing up a hotel in Peshawar. It's very unfortunate, since Bill Phillips and Dave Wineland do great science, but *ARPA does things like this.

BS! IARPA is not cutting off QIS funding - this is not just a shift in investment priorities associated with new leadership. Reading the Nature article we see that Lisa Porter (the new IARPA director) has even touted an increase in (unclassified) support for QIS. There is no issue with congressional support - QIS has gained significant traction in the past years with Congress.

IARPA has decided to take a strict interpretation of the federal acquisition regulations and cease support for other government agencies. While legal, there are also very legal loopholes allowing for justifiable exceptions in the case of special capabilities, etc. NIST clearly falls into this category given the quality and state of research in the field. However, IARPA does not seem interested in having government agencies participate, even if they contain some of the best research groups in the world.

Shortsighted is a nice way of putting it. Insane more appropriately captures my thinking.

Doug says: "Quantum computing, while potentially cool for NSA code-breaking, won't help [the DoD] prevent some fanatics with a truckload of semtex from blowing up a hotel in Peshawar."

That is true only if we take a narrow, short-term point of view about QIT/QIS ... if we take a broader, longer-term point of view, we are led to precisely the opposite conclusion.

Let's adopt the broader point of view that a strategically central (and near-term) payoff of QIT/QIS research is the mathematical framework and software tools for the accurate simulation (with classical resources) of quantum systems.

These QIS/QIT-inspired simulation tools are serving to speed the pace, retire the risks, and optimize the payoff of every enterprise that presses against quantum limits to speed, size, efficiency, sensitivity, and information processing capacity. Which is to say, all new technological enterprises of the 21st century.

Then we are in a position to appreciate that QIT/QIS research (in its broad conception) is one of the most powerful technological tools ever conceived for pursuing (on a generation-long global scale) the strategic goals of Petraeus' Guidance:

http://faculty.washington.edu/sidles/QSEPACK/Kavli/Petraeus_Guidance.pdf

These strategic objectives include specifically: (1) serve the population, (2) live among the people, (5) employ all assets, (6) generate unity of effort, (7) promote reconciliation, (9) foster legitimacy, leadership, and initiative, (10) punch above your weight class, (13) gain nuanced intelligence, (14) interact face-to-face, (15) understand the (technological) neighborhood, (16) build relationships, (17) find sustainable solutions, (18) maintain continuity and tempo, (19) manage expectations, (20) be first with the truth, (21) develop and sustain a narrative that works, (22) live your values, (23) exercise initiative, (24) prepare for and exploit opportunities, (25) learn and adapt.

It is very enlightening to systematically survey the emerging capabilities of QIS/QIT---which already are transformational---in light of the above strategic goals ... a task that amounts to "(21) Develop and sustain a QIS/QIT narrative that works."

As a case study, the December 2008 IEEE Information Theory Society Newsletter as wonderful essays on compressive sampling (by Candes and Tao, and by Donoho) that set forth "a narrative that works" ... and works admirably (in the Petraeus sense). In comparison, the same issue has a essay "Building quantum computers" (by Knill) that is well-written and technically accurate, but does not work as a strategic narrative (in the Petraeus sense).

For example, Knill's summary "To be able to usefully solve problems currently infeasible on classical computers with known quantum algorithms requires thousands of qubits and billions of gates. Although up to eight qubits have been nontrivially manipulated with atomic qubits in ion traps, at this point no one has clearly demonstrated a computationally useful two-qubit register. It is expected that this will be achieved shortly in ion traps."

This summary makes it seem that strategic payoffs from QIT/QIS research are decades away in time, and high-risk in terms of certainty. If QIS/QIT is narrowly, then this is correct.

Yet if QIS/QIT is broadly conceived, the opposite is true: the strategic payoffs from QIS/QIT are right here, right now.

Hi John - I understand what you're saying, and I'm not trying to argue that QIS/QIT isn't important. It's intellectually fascinating stuff that has tremendous potential, and it should be supported. Still, at the risk of a semantic argument when we basically agree on the scientific merit.... I think most people would agree that research into topological quantum computation using the \nu=5/2 quantum Hall system is not directly contributing to short-term DOD strategic goals. And it shouldn't be, except in the very broadest sense. And that's fine! I also think that, QIS/QIT is great stuff, it's premature to declare that it is "one of the most powerful technological tools ever conceived" for combating a counterinsurgency.

"anon" has it right. This is old news and has nothing to do with the level of QIS funding interest at IARPA.

Steve: your guess is as good as mine. "Secrecy and quantum computing" is probably a good fodder for a blog post. :)

John: Who wants to comment when funding is involved? Let the anonymous/pseudonymous comments flow!

Doug ... I think we basically agree ... I will add only (what is my opinion) that coming years will see many people adopting broader views of both QIS/QIT and what you called "counterinsurgency."

SecDef Gates, in a recent essay Beyond Guns and Steel, has characterized the latter task more broadly, as "the development of a sense of responsibility for world order and security."

And here (IMHO) is definitely *is* defensible to assert that QIS/QIT (in its broad conception) is among most powerful technological tools ever conceived for pursuing (on a generation-long global scale) the strategic goals of liberty, democracy, security, and prosperity.

For me (at least) this definitely is "a narrative that works." So much so (in my case), that it is the narrative that leads me to work in QSE/QIT, not the reverse! :)

Hi C: if this is old news would it have made a difference if the letter recently sent to Holdren was sent earlier? The quantum computing community has a real interest in resolving these problems and is, I'm sure, happy to apply whatever pressure it can, politically.

I've heard at least a few experimental quantum computing people complain about their (real or imagined) worries about the stability of funding (i.e. the who's steering the ship problem). I have no idea how to fix this, but it is a concern, and it seems that this

John/Nate: Just a pedantic clarification...this is IARPA which is not DOD. (IARPA reports to the Director of National Intelligence, I think.) :)

anon 6: yep.

The decision by Lisa Porter to not fund other govt agencies makes a lot of sense. It is not insane or shortsighted.

For one thing FFRDCs are not allowed under the law to compete on projects where a private company can supply equivalent technology. If you think no company could produce what NIST / Sandia / etc can produce read up on some cases studies of what happens to basic science when industry thinks there's potential value in it (Celera, GMR, integrated circuits, MRAM, MEMS, microfluidics, etc.)

If the work at NIST has value they will be able to find financing for it. The bigger question is if no-one sees any value in gate model quantum computing (ie no-one wants to fund it) why is this.

On a related note someone above claims that QC has no value to the intelligence community beyond Shor's algorithm and there is no short term value in QC systems. Both are totally false. Discrete optimization problems are EVERYWHERE in intelligence and security applications (anything related to pattern matching, motif detection, automated detection of objects in images, machine learning, etc. etc. etc.) and adiabatic optimization algorithms can be applied to solve these.

The adiabatic quantum optimization algorithms have far more value to this community than Shor's algorithm. The fact that the QC community doesn't appreciate this is centrally related to the problems that QC has in general--in order to properly finance and advance basic science to technology you have to understand and focus on what the end user (Dr. Porter and her peers in this case) really wants.

The decision by Lisa Porter to not fund other govt agencies makes a lot of sense. It is not insane or shortsighted.

Well especially if they're competing with your company ;)

For one thing FFRDCs are not allowed under the law to compete on projects where a private company can supply equivalent technology. If you think no company could produce what NIST / Sandia / etc can produce read up on some cases studies of what happens to basic science when industry thinks there's potential value in it (Celera, GMR, integrated circuits, MRAM, MEMS, microfluidics, etc.)

Well, your first sentence and second sentence don't seem consistent. The second say "could." But what company, exactly, is producing the same technology as, say, the NIST group?

The bigger question is if no-one sees any value in gate model quantum computing (ie no-one wants to fund it) why is this.

Well, as noted, this is not the case...IARPA is increasing funding.

and adiabatic optimization algorithms can be applied to solve these.

They can be applied, but do they really give speedups? That is the $70 million dollar-ish question, isn't it?

The fact that the QC community doesn't appreciate this is centrally related to the problems that QC has in general--in order to properly finance and advance basic science to technology you have to understand and focus on what the end user (Dr. Porter and her peers in this case) really wants.

Well, as Feynman said, "For a successful technology, reality must take precedence over public relations, for Nature cannot be fooled." I'm sure IARPA wants a computer that can solve NP-complete problems in polynomial time, but I don't think Nature gives us that.

(That said, I agree that (1) not enough work has been done on understanding the power of adiabatic quantum algorithms and (2) not enough experimental research has been dedicated to adiabatic quantum technologies. Of course, I'm biased: arXiv:0905.0901.)

[Nits] NIST is an agency, not an FFRDC. My experience with FFRDCs has been less than compelling, since they usually only spend enough time on a problem to develop a basic facility with it. They don't invest. [I wonder if this is a consequence of being non-profit].

Most of the *ARPA R&D investment is driven by the question, "Is there a product in there that I can program and acquire?" [and by 'program' I mean 'turn into something with an industrial disposition']. This really applies to most government R&D.

I should point out we have never competed for US government grant money and have no plans to do so at this point.

Re. companies competing with NIST: the question is not whether they have the same thing sitting on the shelf for funding basic research, it's whether they have the capability to deliver the same "deliverables" at the end of the project. Here are some examples of organizations that likely could if they really wanted to: IBM, HP, Intel, GE, Motorola, ...

The "NP-complete problems in polynomial time" thing is a total red herring. The reality of hardware is that asymptotic scaling arguments are pretty much meaningless at this point.

No classical methods perform this well. Are you arguing that solving discrete optimization problems with classical algorithms (with all of their flaws and drawbacks) has no value?

AQO is nothing more or less than a new class of algorithms for solving discrete optimization problems. One organization can have the unique capability to run them (they require unique hardware), which is intriguing from the business perspective.

It will turn out that these systems will be very good on some types of instance classes and they will suck on others, just like all solvers of these types of problems. Our simulations of the hardware show pre-factor speed-ups coming from the special-purpose nature of the (superconducting) chips of 6-7 orders of magnitude over CPLEX running on a high end server for problems ~ 128 variables. So regardless of eventual scaling arguments these chips are going to do what they're going to do extremely quickly.

With apologies to Feynman, successful technologies are ones that get built and sold, with profits from sales feeding back into R&D, leading to new products that get built and sold.

> Do you think this is a way to twist their arms to go
> "underground" and work on classified stuff?

Steve, I think you give the government too much credit. This simply strikes me as a typical bureaucratic move akin to the FHA requiring me to paint my house before they'd give me a loan to fix my furnace - following the letter of the law to its extreme.

I will say that it is sad that this was the biggest news to come out of the recent meeting in Virginia about a national QIS initiative. Be on the lookout for the next issue of The Quantum Times (hopefully coming out this week) where Mark Wilde and Andrew Landahl have a nice article about this.

Geordie,

"The bigger question is if no-one sees any value in gate model quantum computing (ie no-one wants to fund it) why is this."

What are you talking about? Not everything in the world is about DWave or your marketing perspective.

IARPA has given NO indication that it favors adiabatic QC over the gate model, or that it doesn't want to fund quantum information so your arguments are irrelevant. Maybe AQC is the way to go, who knows, but that's not what's under discussion. Having been at the IARPA meetings, I can tell you that this is not some secret coup to fund AQC over gate-model QC.

NIST (Wineland) was the first group to demonstrate a convincing coherent quantum gate. They have pioneered experiments in quantum error correction, quantum algorithms, entanglement, and teleportation. There is not a single other group in the world that has demonstrated these capabilities, or the consistent advancements the NIST group has shown.

Indeed, FAVORING private industry over government agencies helps ensure competitiveness in the private sector. That's fine. But the FAR also has exceptions for the specific case when similar capabilities are not present or reasonably available outside of the government. Given the NIST track record, I think it's nearly impossible to suggest that similar capabilities are readily available elsewhere. And before someone objects with Rainer Blatt and Chris Monroe (both top ion groups), there remains plenty of room for NIST to contribute - otherwise there wouldn't be new NIST articles in Nature every other week.

As such, it becomes illogical to arbitrarily cut off funding to government agencies such as NIST.

anon: Sorry what I meant wasn't clear in my last post. I didn't mean to single out gate model vs. AQC and it is certainly true that no-one in funding circles favors AQC over gate model, nor should they.

What I meant is that the level of funding overall for *all* approaches to QC is very low compared to where it should be. Let's be honest, a single M1 tank costs more than the sum total of QC funding over all research groups in the US last year. The benefits of having quantum computers are not being quantified properly to the people who decide what gets funded.

There has to be a clearly and simply articulated argument for what the value of quantum computation is in dollars over time. If the number is big and enough people buy the argument then you can invest a lot into it even if the risk is high and the times are long.

I guess my central point is that programs like those at NIST should be getting WAY more funding for far more ambitious projects. One of the reasons they're not is that the connection of the technology to the end user and the benefits that end user will see are not being quantified.

I don't see this as a marketing perspective and it isn't about D-Wave. It's about what unlocks the really big sources of financing required (in my view $100M+) for any organization to make real progress on these horribly difficult projects.

Geordie,

I appreciate your response. I, too, agree that bigger funding pots are appropriate given the potential impact QIS can have on technology of interest to the government.

I disagree, however, with your comment that one reason NIST is not getting more funding is that the end-user benefits are not being quantified. Inside funding circles there is a ton of emphasis on potential returns for all kinds of investment in quantum information - gate model QC, AQC, topological QC, QKD, metrology, etc.

People in government seem to get it about how important QIS might be down the road. This is again evidenced by the manner in which Lisa Porter defended herself - significant expected increases in QIS funding.

The issue remains the same: a bureaucratic excuse is being used to cut-off funding for one of the most prolific experimental research groups in the field. That is a travesty.

I don't mind mentioning that Ian has a well-reasoned (IMHO) essay Why a national initiative in quantum information is needed posted on his weblog Quantum Moxie.

I should point out we have never competed for US government grant money and have no plans to do so at this point.

Which doesn't mean that those who have received such money aren't your competitors (cue the random commenter who berates me every time I say that anything compares to D-wave!)

No classical methods perform this well. Are you arguing that solving discrete optimization problems with classical algorithms (with all of their flaws and drawbacks) has no value?

Of course not. The way you phrased your sentence makes it sound like it is always possible to solve the problems that confront a customer. I was trying to point this out in an exaggerated way.

It will turn out that these systems will be very good on some types of instance classes and they will suck on others, just like all solvers of these types of problems. Our simulations of the hardware show pre-factor speed-ups coming from the special-purpose nature of the (superconducting) chips of 6-7 orders of magnitude over CPLEX running on a high end server for problems ~ 128 variables. So regardless of eventual scaling arguments these chips are going to do what they're going to do extremely quickly.

With noise? I assume this is a quadratic optimization problem? Why compare to a high end server and not something more beefy? Inquiring minds want to know (okay just this loafer.)

With apologies to Feynman, successful technologies are ones that get built and sold, with profits from sales feeding back into R&D, leading to new products that get built and sold.

Indeed, but Nature still sets the prereqs.

Ian: I think your libertarian leanings predestine you to think the government couldn't pull of something like this in secret :)

Two things to add:

First, can anyone post a copy of the letter? Maybe one of the 12 that signed it? I'd like to know what the Nature article was specifically talking about.

Second, in regards to IARPA's position on FFRDC's et al., the current open BAA states explicitly:
"Other Government Agencies, Federally Funded Research and Development Centers (FFRDCs), University Affiliated Research Centers (UARCs), and any other similar type of organization that has a special relationship with the Government...are not eligible to submit proposals under this BAA or participate as team members under proposals submitted by eligible entities."

More specifically, the previous Coherent Superconducting Qubit BAA from IARPA used the following language:
"In accordance with FAR 35.017 (a) (2), other government agencies and Federally Funded Research and Development Centers (FFRDCs) are not eligible to submit proposals under this BAA or participate as team members under proposals submitted by eligible entities."

In my interpretation, however, FAR 35.107 does not exclude FFRDC's from use by a non-sponsoring agency but rather explicitly endorses the idea (FAR 35.107-3(b)) while FAR 17.503 defines the threshold for justification.

My point in noting these Federal Acquistion Regulations is that IARPA, or any government agency, is bound by law as to how they administer the monies provided to them by congressional authorizations. How Porter has dodged these regulation and the expected oversight is what I don't get....