Research Team in Japan Reports Excess Heat - (Nissan Motors among otheres)

Below is a photo of Takahashi's equipment, from:

http://lenr-canr.org/acrobat/TakahashiAphenomenol.pdf

Anyone who imagines you could "send" this equipment to MIT or Caltech and have someone there operate it has not read the literature and knows nothing about how these experiments are performed.

Quote from Shane D.

Is it good to have so many variations? Seems to me, that if one model was agreed upon by all to be the best, then all used it, everyone would at least be starting on the same page. But that is just my layman's perspective.

I think it is better to have many variations, for the following reasons:

If they all used the same type of calorimeter, people might suspect they are all making the same systematic error. Since the systems are different there cannot be one systematic error.

Different types of calorimeters have different strengths and weaknesses.

Some are well-suited to one kind of experiment. For example, Melvin Miles' equipment worked well with the method of helium collection that he decided to use.

Some calorimeters are very expensive, such as the one at SRI. Most researchers could not afford them, but it is a good thing that a few researchers could afford them.

" ±1.5％error " is probably necessary when you are looking for excess heat of 5-10% of input

as the Kobe group reported earlier on

Once you have excess heat running in the 50 to 100% level less accuracy is needed

such as with air calorimetry, as Mizuno used,which is cheaper and more flexible

to accommodate different reactor configurations.

If and when economic levels of excess heat of 200% are reached

where the durability, controllability ,nanopreparation are needing to be optimised,

other assays of LENR activity e.g Helium/Xenon/He-3 /nitrogen levels may be more informative

rather than just the temperature sensors.

Massa Mario "This means that they do not believe they have had sufficient experimental results to declare that they really exist. "

The researchers call their results modest in the ICCF20 report

Massa Mario tends to exaggerate for rhetorical effect, as with asserting that Mizuno does not know how to calculate circumference.

I fear that his website is not well visited so perhaps he feels the need to shout out some sensation to get attention

Science is all about verification and measurement. Also the Kobe researchers Kitamura et al are trying to find alloys which give a better effect.

Apparently a mixture of palladium and nickel in a 1:10 ratio has been found to give the best Xs heat.

"The maximum excess power is about 10 W in the PNZ3#3-3 and #3-5 phases with the input power of (W1, W2) = (94 W, 40 W)"

The researchers know as well as you and I and Mario that a ~10% xs heat to input ratio is useless practically which is why they are proposing 5 years to develop some kind of industrial application which might give then some funding leeway to get up to 200% xs heat level somehow.

It's no use saying three years and getting only two years funding.. I'd go for more.

Nissan and Toyota probably have plenty of competing uses for the small amount of yen they are donating.

Quote from JedRothwell

If they all used the same type of calorimeter, people might suspect they are all making the same systematic error.

Exactly. And the data analysis methodology is an integral part of the calorimetric method, just as the construction of the equipment is. All cold fusion calorimetrists use a lumped parameter approach to their data analysis, Irregardless of their design (some designs such as single point measurements as in isoperibolic calorimetry can't help this, there is no other option), and this includes Takahashi, et al. Thus they are all doing what Jed clearly identifies as a poor method. The papers put out by Takahashi covering the work in the recent report are another example of this, and what's worse is they show that their system is highly heterogeneous. Yet they still use the lumped parameter approach (i.e. 1 equation to describe the calorimeter), which implicitly assumes that there are no significant temperature variations. Thus they are set up for a CCS problem. Their calorimeter captures about 75% or so of the input heat (as opposed to 98-99% levels reached by Storms and McKubre) which makes the potential impact of a CCS larger.

Further, they don't do standard error analysis (Propagation of Error) to establish their accuracy and precision, and they fail to treat the calibration constants as experimentally determined numbers (which they are), so their error band calculations are incorrect by normal standards.

All in all, not a very good situation to claim extraordinary results from.

I will also note that Takahashi, et al claim they were trying to duplicate the Kitamura, et al Physics Letters A paper that I tried to publish a Comment on. The manuscript for that Comment is included in my previously mentioned whitepaper. In it I show that those results were consistent with prior research with those materials and thus point out a non-extraordinary explanation for the observations. If Takahashi is replicating Kitamura, my explanation applies to their work also.

Kirkshanahan

"they don't do standard error analysis (Propagation of Error) to establish their accuracy and precision"

This appears to be 100% assertion on Kirkshanahan's part.

I did propagation of error in high school physics experiments when the Vietnam war was young.

Taking account of the overall system errors by considering component contributions is standard

Where does it state that Takahashi et al have specifically ignored propagation of error in the 2016 report?

Here . http://vixra.org/abs/1612.0250

or here

" Evaluated accuracy of the calorimetry system is satisfactory, namely less than ±1.5％error in thermal-power measurement, less than ±0.1℃ error in temperature detection by thermos-couples and RTDs, and less than ±2％ error in thermal flux measurement. "

Anomalous Heat Effects by Interaction of Nano-Metals and H(D)-Gas

Authors: A. Takahashi, A. Kitamura, K. Takahashi, R. Seto, T. Yokose, A. Taniike, Y. Furuyama

“This appears to be 100% assertion on Kirkshanahan's part.”

You’re not very good at discerning things are you? You should actually read the relevant papers and such.

“I did propagation of error in high school physics experiments when the Vietnam war was young. Taking account of the overall system errors by considering component contributions is standard”

Good for you. Yes, POE is a standard technique as I said. You can check here for a description (https://en.wikipedia.org/wiki/Propagation_of_uncertainty), but it has been around for a long time. I have a book by Hugh D. Young (“Statistical Treatment of Experimental Data”) written in 1962 that discusses it on pages 3-9 and 96-101. Any good text on the subject will cover it. Which is why it is so sad that these CF researchers don’t do it right.

“Where does it state that Takahashi et al have specifically ignored propagation of error in the 2016 report?”

Y. Iwamura et al. / Journal of Condensed Matter Nuclear Science 24 (2017) 191–201

Page 195 has the relevant info. They use the approximate equation

δ (H_EX ) ≈ | δ (F_R )| ρ C ΔT / ɳ + | δ (ΔT)|    F_R ρ C / ɳ + | δ (W)| ( Their eqn. (3).)

(Hopefully the Greek letters and symbols will show up properly in the forum post.)

to compute the error instead of the POE equation. Note that their ‘δ’ is some sort of approximation to the normal standard deviation, σ. It is mathematically incorrect to add sigmas, because it is variance, or σ², that is additive.

They also state:

“Considering that experimental variables are F_R, ΔT and W, we can assume that error range of the calculated excess heat is the sum of fluctuations of oil flow rate, temperature difference and input electrical.”

Note the word “assume” and the lack of ɳ, the ‘calibration constant’, being listed as a variable (which it is as it is experimentally determined).

I did some back-of-the-envelope calcs and the ɳ term can be significant. I also checked up on their use of density and heat capacity and found that they may be oversimplifying as well. See:

“A Mass-Flow-Calorimetry System for Scaled-up Experiments on Anomalous Heat Evolution at Elevated Temperatures” A. Kitamura, A. Takahashi, R. Seto, Y. Fujita, A. Taniike and Y. Furuyama, J. Condensed Matter Nucl. Sci. 15 (2014) 1–9

Where you will find figure 2 to show those quantities as functions of temperature, which convinced me that using an average temperature in their calibration equation could also add 10-20% errors in.

All in all, very sloppy error analysis. My B-O-E calcs led me to believe 3sigma levels of 20-30% of input power are easily obtained, and since we are talking about radical new physics here if LENR is real, we probably should use the 5sigma level just like they used in searching for the Higgs boson, which give roughly 30-75% of input power for the error bars. None of the presented results are out of that range I believe.

And don't miss that _here_ I _am_ talking about random error. The heterogeneous temperature distributions however can potentially allow for a systematic error as well (the 'CCS'), just like in the case I examined in my 2002 publication on the reanalyzed Storms data, on top of the random error.

Kirkshanahan "they don't do standard error analysis"

You suggest that the errors in ɳ and the averaging of ΔT , C and ρ contribute large errors that the Iwamura et al have ignored.

You write such things " could also add 10-20% errors in."

Does that mean 10-20% of their calculated error figure or 10-20% of the actual calculated quantity?

For example Iwamura et al get 0.26W error for 80W input (with an excess heat rate of 5W)

Are you suggesting that the HEX error is up to 0.26 +20%= 0.31W???

or =+20% of 5W = +1W??.

In the latter case this would mean your error calc is 5 times bigger than that of the Iwamura et al .

This is an extraordinary difference.

You wrote B-O-E . What values of ɳ , ΔT , C and ρ are you plugging into your B-O-E?

For the 80W input energy values Iwamura et al have written error estimates.

oil flowrate error = 0:012(ml/min);

oil temperature change error = 0:261(K);

input energy rate error = 0:031(W);

excess heat energy rate error (HEX) = 0:260 (W):

What are your estimates for these?

Can someone share the full paper even if it's in Japanese? Thanks.

http://www.google.com.au/url?s…Vaw1pmgmF4z841xzU1LgSvhbP

Pg 191 and following Journal JCMNS Volume24

Quote from bocijn

http://www.google.com.au/url?s…Vaw1pmgmF4z841xzU1LgSvhbP

Pg 191 and following Journal JCMNS Volume24

Thank you kind Sir.

That's better.

This would be a useful example (of no binding significance, but still helpful) to investigate the way that systematic errors might, or might not, be significant in such CF experiments.

Many temperature measurement points enable us to judge whether observed excess anomalous heat is real or not.

From p192. We'd therefore hope that errors due to different temperature distribution inside the reactor could be detected or eliminated. But, I won't assume that is the case!

Also, there may be any number of approximations in the calculations given there. But, if such approximations are truly conservative I see little harm. I'd still rather more precision because a precisely measured quantity can sometimes indicate an anomaly that otherwise can be overlooked, and might indicate an methodological issue brushed under the carpet.

Error bound calculation. In this case summing variances instead of root mean squaring them is conservative, and unproblematic, when establishing error bounds. Kirk: as you know I often find your posts here invaluable: you bring facts and analysis to the table that others overlook. You might properly point out that such over-approximation of error bounds (especially without note) indicates a possibly cavalier approach to statistical analysis. But I can't see anyway that it harms the subsequent analysis except to weaken slightly the experimental results. The more serious issue in the given calculation is that it has no allowance for change in efficiency due to differences in reactor conditions between calibration runs and real ones. A better error calculation would note that sensitivity and therefore allow CCS to be eliminated or detected.

Other errors. Kirk implies there are other errors in the Iwamura analysis. I can't see that a presupposition that there are (Kirk, perhaps) or are not (Bocjin, perhaps) without careful examination does anyone any good. As we say with the FTL neutrino measurements, when results are anomalous everyone properly expects it likely there is some subtle or obvious error underlying the analysis, and they are usually right.

I'm a bit pressed for time now, but I find all these Japanese experiments interesting in that they have multi-lab replication (but of course that, from the writeups so far, does not rule out flawed and replicated methodology) and claim results that are strong given their expensive calorimetry. However, purely from the outside, I note the high temperature and therefore higher heat losees (Kirk mentions 75%ish efficiency). At lower efficiencies anything that changes the efficiency and therefore the calibration factor is more significant. this is therefore a prime candidate for CCS-style effects.

I personally like things to be understood. These Japanese results are well enough documented now that an external observer could look at them and note flaws to be addressed, or not. It looks like they have enough support to address any flaws.

In the process of doing this a better understanding will out. Either real flaws remain unaddressed, or real anomalous behaviour remains unremarked by the outside world.

To calculate the error (a value-laden term, ‘standard deviation’ is more correct and usual, but the term ‘error’ is ubiquitous) in a quantity computed from experimental variables, the Propagation of Error (POE) or Propagation of Uncertainty formula is used. In the following I will not be using Greek letters, I will use the standard English ones instead. I here define:

in general, x_i is used to indicate ‘x subscript i’

s = standard deviation, as per the normal (or random) distribution equation

p(x) = partial of x, or with respect to x, i.e. p(R)/p(x) is the partial derivative of the function R with respect to x

n = 1/eta = the reciprocal of the heat capture efficiency as defined in the paper referenced, note that in other papers they sometimes call eta Rh (below I use the reciprocal for simplicity to make the math easier to understand), it is also the ‘calibration constant’ and is determined by ‘calibrating’

SUM [ ] – indicates a summation of the terms inside the brackets

r = rho – here meaning the density

C_p = C sub p – the heat capacity or specific heat of the calorimeter fluid

dT = delta T = a temperature difference

The POE formula is, given a function R that is a function of several experimental variables (x_i), the variance in R is computed from the summation of terms composed of the partial derivative of R with respect to x_i squared times the standard deviation of the variable x_i squared, i.e.

(s_R)^2 = SUM[ (p(R)/p(x_i))^2 * s_x_i^2 ]

In our case R = P_out = n * F * r * C_p * dT, so the partials are: p(r)/p(n) = F * r * C_p * dT, p(r)/p(F) = n * r * C_p * dt, p(r)/p(r) = n * F * C_p * Dt, p(r)/p(C_p) = n * F * r * dT, p(r)/p(dT) = n * F * r * C_p

So the standard deviation of Pout = the sum of the squares of those terms multiplied by the variables observed variance (variance is the standard deviation squared).

Now a little simplifying trick. In simple equations such as this one can divide the summation by the square of the function and each term will then convert to the variance of variable x_i divided by x_i squared, which equals the standard deviation divided by the variable, all squared. This will then allow one to talk about uncertainty using fractions of the variables (or percentages if desired).

So, we get

s_R^2/Pout^2 = ( s_R/ R )^2 = SUM [ s_x_i^2/x_i^2 ] = SUM [ (s_x_i / x_i)^2 ] (‘R’ = Pout)

It is now easy to see that variables with small fractional variation will contribute little to the overall sum.

Takahashi et al only focus on F and dT, as noted by bocijn. F is not specified in the paper so we will have to start assuming, but for consistency with the authors assertions, let’s assume the fraction s_F/F is about .01 (or 1%). That squared is .0001. It may be less than that, but we don’t know for sure. dT is given as .26 degrees. Shall we use 300C = 573 K? That means the ratio is .26/573 = .00045, squared is 2.06e -7, so if we only consider these two variables as do the authors, the fractional error in Pout is very small (~1%).

How about F and p? The authors ignore them. We can compute those values from the equations on Figure 2 in the paper J. Condensed Matter Nucl. Sci. 15 (2014) 1–9 for the temperature range that the authors cite (200-300C), and we get 2.18 <= C_p <= 2.52 J/g^oK and 852 <= p <= 921 kg/m^3, for ranges of 0.34 and 69 or fractions (based on smallest numbers) of 0.13 and .081 or as percentages 13% and 8%. That is actually a significant span. In their equations the authors use the average temperature, but they don’t define what that is based on, the 200-300C range or a 25-300C range (which would be much worse than the numbers given here). So depending on the temperature variation of the experiment, one could get standard deviations in the excess power on the order of 10-15%.

Now we have to decide how we will decide if we are ‘out of the noise’ or not. Do we use the standard 3 sigma, or do we get tougher due the import and unlikelihood of the results and use 5 sigma like the Higgs boson hunters? 3 sigma is 30-45%, 5 sigma is 50-75%. So on an input power of 134W, that gives a noise band of +/- 40-60W or 67-100W. The authors report: “The peak ratios of excess heat to input power were about 4% and 5% for 80 and 134 W input, respectively.”

To summarize, using an average T in the excess heat calculation greatly increases the error and the assumption that density and specific heat are of no import becomes wrong. The reported results are well within a reasonable noise band. This could be fixed by not doing that of course, but that’s not what they say they did.

Note that I haven’t talked about n yet. So what about it? It is certainly an experimental variable. It is determined via an experimental process. So it should be evaluated as to its contribution to the noise band on excess heat. But, once again, we have no information on the precision, so we have to resort to ‘sensitivity analysis’ to try to decide if possible errors in it are important. Having converted to the fractional approach for computing Pout's error, we can simply state that it will have a directly proportionate effect. If n's standard deviation is 1% of its value, that will cover the 4-5% value reported in the quote above. If it is bigger it will allow for larger errors, and those potential results would be deemed ‘in the noise’ as well.

All of the above is standard error computation based on random statistics. Because the temperature distribution inside the calorimeter are quite large, as shown in the Figures, the possibility of a CCS exists. One would need to replicate the experiment many times, specifically including replication of the calorimeter/cell assembly, which is likely where the nonuniforminty is first introduced. Running multiple runs on 1 batch of powder would not even touch this issue. The other way to do it is the same as I suggested for the F&P-type cells. Put at least two resistive heaters in the powder and run them at different power levels to see if the calibration holds when the temperature distribution is changed.

Kirkshanahan

Are you suggesting that the HEX error is up to 0.26 +20%= 0.31W???

or =+20% of 5W = +1W??.

In the latter case this would mean your error calc is 5 times bigger than that of the Iwamura et al .

This is an extraordinary difference.

You wrote B-O-E . What values of ɳ , ΔT , C and ρ are you plugging into your B-O-E?

For the 80W input energy values Iwamura et al have written error estimates.

oil flowrate error = 0:012(ml/min);

oil temperature change error = 0:261(K);

input energy rate error = 0:031(W);

excess heat energy rate error (HEX) = 0:260 (W):

What are your estimates

oil flowrate error = ? (ml/min);

oil temperature change error =? (K);

input energy rate error = ? W);

excess heat energy rate error (HEX) = ? (W):