Chapter 1: Preliminary Description of Error Analysis

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement not replaced the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

If you make an error in everyday life, you know, you basically screwed up.

Right.

You blundered, you took a wrong turn, or sent an email to the completely wrong person.

Exactly.

But if you make an error in a college physics lab, you haven't actually made a mistake.

You are just, you know, confronting the inescapable uncertainty of the physical universe.

Yeah, which is a pretty huge mindset shift.

Today we are basically shattering the illusion of mathematical perfection.

Oh, for sure.

So welcome to this special one -on -one tutoring deep dive.

If you're prepping for your first real encounter with error analysis, you are exactly where you need to be.

Absolutely.

Our mission today is to master the foundational concepts of error analysis.

We're moving step by step from basic definitions straight into practical experimental reasoning.

Right.

We're taking the core lessons from chapter one of introduction to error analysis and, you know, translating them into the exact skills you're going to need at your lab bench tomorrow morning.

Because that fundamental shift in how we define the word error is, well, it's the very first hurdle every science student has to clear.

Yeah, I mean, if you walk into the lab thinking of an error as a mistake,

your instinct is going to be to try and perfectly eliminate it, right?

Exactly.

You'll obsess over getting the quote unquote right answer.

But once you realize that error simply means uncertainty,

your entire objective changes.

Because your job isn't to magically make that uncertainty disappear.

Right.

Your job is to realistically estimate it.

And we're going to walk you through how to evaluate these uncertainties using just plain common sense.

Yeah, you really don't need advanced calculus to build a rock -solid foundation for all your future experimental work.

Not at all.

But before we can calculate uncertainty, we have to understand why it's just physically impossible to avoid it in the first place.

Okay, so let's put you in the shoes of a carpenter.

Say you're trying to measure the height of a doorway.

You look at the door and make a crude visual estimate, right?

Like 210 centimeters.

Yeah.

And obviously there's a massive amount of blurriness there.

Oh, yeah.

If pressed, you'd probably admit the real height could realistically be anywhere between, I don't know, 205 and 215 centimeters.

Right, so your immediate instinct as someone trying to do good work is to fix that blurriness.

So you grab a standard tape measure.

Makes sense.

You hook it to the floor, pull it up to the top of the frame, and now you find the height is exactly 211 .3 centimeters.

Okay, so you're definitely more precise now.

You are.

But look closely at the physical reality of that tape measure.

Let's say it's graduated in half centimeters.

The top of the door probably doesn't line up perfectly with one of those little marks.

You have to use your eyes to estimate where it falls in the empty space between them.

Oh, I see.

And even if it does seem to line up perfectly, the printed ink mark on the tape measure has an actual physical width, right?

Exactly.

It might be a millimeter wide itself.

Yeah.

So you still have to estimate where the exact edge of the door lies within that literal printed line of ink.

That is so frustrating.

So, okay, let's say you throw away the tape measure.

You become obsessively determined to find the true height of this door.

You go out and buy an insanely expensive laser interferometer.

You're now bouncing light beams off the door frame, measuring distances down to the wavelength of light.

Which is what, about 0 .5 times 10 to the negative sixth meters?

Right.

Something crazy small like that.

Your precision is microscopic now, but here's the kicker.

You still do not know the exact height of the doorway.

And this brings us to a really foundational scientific concept.

It's known as the problem of definition.

The problem of definition.

Okay.

Explain that.

Well, as you strive for this microscopic precision,

physical reality actually starts getting in the way of your math.

A wooden doorway isn't a static, perfect, geometric object existing in some vacuum.

Right.

It's made of actual wood fibers.

Exactly.

If the temperature in the room goes up by a single degree, the wood expands and the height changes.

Oh wow.

If the humidity changes, the wood absorbs moisture and swells.

The height changes again.

If you accidentally rub off a microscopic layer of dirt or paint while setting up your laser.

The height changes again.

Yes.

There is literally no such thing as the quote,

exact height of the doorway.

The quantity itself just is not perfectly well defined in nature.

That is so wild.

It's entirely like zooming in on a digital photograph.

Oh, that's a great way to put it.

Right.

You keep zooming in, assuming you're going to uncover the true underlying details of the image, but eventually you don't find the truth.

You just hit pixelation.

Exactly.

The definition of the image literally breaks down into these blurry squares.

Yeah.

And the physical world basically has its own kind of pixelation.

It really does.

Now, in everyday life, we don't bother stating these uncertainties because they just don't matter to the task at hand.

Right.

If a doorway is a microscopic fraction of a millimeter off, the door still shuts.

The carpenter gets paid.

Right.

But in scientific measurements,

stating your uncertainty is absolutely mandatory because knowing the spread of your uncertainty is often the only thing that separates a completely useless experiment from a major scientific breakthrough.

Okay, wait, let's dig into that.

If every single measurement we ever take is inherently blurry, how do we actually use data to prove anything?

That's the big question.

Let's look at a classic historical problem to see how this plays out in practice.

Imagine you're tasked with solving the Archimedes crown problem.

Okay, classic.

Right.

So a king hands you a crown.

You need to find out if it's made of pure 18 karat gold, which has a known established density of 15 .5 grams per cubic centimeter.

Or if it's a cheaper suspected alloy, which has a density of 13 .8 grams per cubic centimeter.

Exactly.

To figure this out, you basically just need to measure the density of the crown and compare it to those two known numbers.

So let's hire two experts to measure it, George and Martha.

This is going to demonstrate how measurement data is actually interpreted in the real world.

Okay, let's visualize their findings on a graph.

Like picture the vertical axis representing density.

George goes first.

Right.

And George uses a pretty crude water displacement method.

He dunks the crown in a beaker, splashes some water on the table, and reads the volume on this chief faded graduated cylinder.

Real messy work.

Very messy.

So George reports his best estimate for the density is 15.

On our imaginary graph, that's a dot right at 15.

Okay, got it.

But because his physical methods were so sloppy, he has to include vertical error bars showing his probable range of uncertainty.

He admits the true density is realistically anywhere between 13 .5 and 16 .5.

Wow.

Okay.

Then we have our second expert, Martha.

And Martha is meticulous.

Very.

She uses a precise mechanical balance to find the mass and a carefully calibrated overflow vessel to capture every single drop of displaced water.

She reports a best estimate of 13 .9.

So her dot is at 13 .9.

Right.

And her error bars are incredibly tight.

Her probable range is just between 13 .7 and 14 .1.

Okay, so let's analyze what these two different sets of data actually tell us.

Look at George's range.

It spans from 13 .5 all the way to 16 .5.

Right.

That's a massive window.

Exactly.

And that massive window captures 15 .5, which is the density of pure gold.

But it also captures 13 .8, the density of the cheap alloy.

Both possibilities easily fit within his margins of error.

Wait, so George's math isn't actually wrong, right?

Like, the true density probably does fall somewhere in his massive window.

Oh, his math is fine.

His measurement just doesn't help us solve the problem at all.

In fact, if we ignored his uncertainty and only looked at his best estimate of 15, we might be completely misled.

We'd think the crown is genuine gold, simply because 15 is numerically closer to 15 .5.

Yeah, that is the danger of ignoring the spread.

We can draw absolutely no conclusion from George's work.

None.

But Martha's measurement is what actually saves the day.

Her tight range of 13 .7 to 14 .1 comfortably includes the cheap alloy at 13 .8.

Right.

But much more importantly, it completely excludes the pure gold at 15 .5.

Because her precisely stated uncertainty does not overlap with the gold, she definitively proves the crown is a fake.

Boom.

And this proves why you, as a student stepping into a lab, can never just assert an uncertainty in a lab report without justifying exactly how you got it.

Our entire decision to accuse the king's jeweler of fraud hinges on Martha's claim that her uncertainty is strictly between 13 .7 and 14 .1.

Right.

She has to give a sufficient reason to believe her claim based on exactly how she conducted the experiment.

An unjustified uncertainty is just numbers on a page.

It is basically as useless as no uncertainty at all.

Exactly.

And, you know, this rigorous justification isn't just for catching crooked jewelers.

It is literally the backbone of high -stakes engineering.

Oh, for sure.

When engineers design the containment vessels for nuclear power plants, they need to know the exact failure limits of the materials.

When they calculate braking distances for high -speed trains, they're dealing with massive amounts of kinetic energy.

Right.

Failure to do error analysis correctly in those fields literally costs lives.

It does.

And it also dictates our fundamental understanding of the universe.

In the basic sciences, whenever a new radical theory is proposed, it must be tested against older accepted theories.

And because of the inevitable experimental uncertainties we've been discussing, the experimental results have to be analyzed incredibly carefully.

Right.

You have to keep refining it until the data definitively singles out one theory over the other, just like Martha with the crown.

Speaking of which, a legendary example of this kind of high -stakes showdown is the 1919 test of Einstein's theory of general relativity.

Oh, this is such a great example.

Right.

So in 1916, Einstein predicts that light from a distant star will physically bend as it passes near the immense gravity of our sun.

Specifically, his math predicted a bending angle of 1 .8 seconds of arc.

But the old accepted classical physics predicted something entirely different.

Classical theory said there would either be no bending at all, an angle of zero, or a very slight bending of just 0 .9 seconds of arc.

So the scientific community suddenly had a clear defining test.

Observe a star perfectly aligned with the edge of the sun and measure the bending angle.

Right.

If it's 1 .8, Einstein rewrites physics.

If it's 0 .9, classical theory holds.

But measuring starlight passing the sun is physically grueling.

You can really only do it during a total solar eclipse, when the moon blocks the sun's glare.

Right.

And think about the conditions during an eclipse.

The temperature drops rapidly, the metal of the telescopes actually contracts, the glass lenses slightly warp.

And the Earth's atmosphere distorts the light anyway.

Exactly.

The physical uncertainties are just immense.

Despite all that, a team of scientists Dyson, Eddington, and Davidson went out in 1919 and actually made the measurement.

And what did they find?

Their best estimate for the bending angle was 2 seconds of arc.

But as we just learned from George and Martha, the best estimate isn't enough to dethrone classical physics.

No.

They had to calculate their spread.

And they reported with 95 % confidence that the true value lay between 1 .7 and 2 .3 seconds.

Okay.

So look at how those numbers align.

Einstein's prediction of 1 .8 falls perfectly inside their stated uncertainty range of 1 .7 to 2 .3.

Right.

And the classical predictions of 0 and 0 .9 fall completely outside of it.

Exactly.

The result was consistent with relativity and inconsistent with the older theories.

It totally vindicated Einstein.

But you know, at the time,

people argued fiercely about it.

Oh, I bet.

Critics suggested the experimenters had badly underestimated their uncertainties.

Wait, so the entire scientific debate wasn't even about the measurement itself.

Like nobody argued they didn't measure 2 seconds.

Nope.

The fight was purely about the error bars.

Wow.

If Eddington's error bars were actually wider due to the temperature shifts on the telescope, say his true range was actually between 0 .5 and 3 .5, the whole experiment would have been a George situation.

Oh, right.

Because both Einstein's 1 .8 and Newton's 0 .9 would have fit inside that wider window.

Exactly.

It would have proven absolutely nothing.

Their ability to reliably estimate their uncertainties in the face of harsh physical conditions is literally what changed physics forever.

That's incredible.

But, you know, bringing this back to your introductory physics laboratory, you might feel a bit disconnected from Eddington proving Einstein right?

Yeah, measuring a ball bearing is a bit different than an eclipse.

Right.

When your professor asks you to measure the acceleration of gravity, Jury, by dropping a steel ball bearing, you already know the accepted answer is about 9 .8 meters per second squared.

Right.

It can feel like a totally artificial, pointless exercise to test a constant that has been proven millions of times over.

Honestly, dropping a ball bearing to find G always felt like busy work to me.

I could just open Wikipedia and get a more accurate number than my little stopwatch could ever provide.

And that's exactly what everyone thinks.

Yeah.

But the point of the lab is not to discover the value of G.

It's not.

No.

The point is to practice the exact same error analysis skills that Eddington and Martha used.

It is a training ground for estimating uncertainty with the flawed physical equipment sitting right there on your desk.

Oh, I love that.

Once you realize the goal is mapping the boundaries of your own uncertainty, the experiment becomes an incredibly instructive challenge.

Exactly.

So we've talked about why we need these boundaries.

Now let's talk about how we actually find them in the lab.

Right.

Getting practical.

And it gets incredibly practical when you're just reading simple scales.

Sometimes plain old common sense is your absolute best statistical tool.

Okay.

Imagine measuring a pencil with a standard plastic millimeter ruler.

You place the flat end of the pencil at the zero mark.

Then you look at the sharpened tip.

Let's say the tip comes closest to the 36 millimeter mark.

Right.

So your best estimate for the length is clearly 36 millimeters.

Yeah.

And the physical marks on the ruler are one millimeter apart.

Using basic visual judgment, you can reasonably decide that the length is undoubtedly closer to 36 than it is to 35 or 37.

Yeah.

You really couldn't realistically read it any more precisely than that with your naked eye anyway.

Exactly.

Therefore, your best estimate is 36 millimeters and your probable range is half a millimeter in either direction.

So 35 .5 to 36 .5 millimeters you've measured to the nearest millimeter.

Now there is a vital hidden convention here that every science student needs to memorize right now.

Yes.

Listen to this.

If a scientist writes down length equals 36 millimeters without stating any specific plus or minus uncertainty,

the universal scientific convention is to presume they mean the value lies closer to 36 than to 35 or 37.

Right.

The stated number automatically implies a range of 35 .5 to 36 .5.

Yes.

And this convention is exactly why relying on a pocket calculator without thinking is so incredibly dangerous in a lab setting.

Oh, this is the biggest trap.

When you divide two numbers in your calculator to find like an average or density, the digital screen is going to spit out something like 123 .456.

Right.

Please, whatever you do, do not simply copy that entire string of numbers into your lab report.

Because if you write down 123 .456 millimeters unthinkingly, you are invoking that scientific convention we just talked about.

Yep.

The reader is entirely entitled to assume your number is definitely correct to six significant figures.

You are essentially implying an uncertainty of 0 .00005 millimeters.

You are literally telling your professor that you measured a standard wooden pencil with a plastic ruler and you know its length down to the width of a single bacterium.

Which is physically absurd.

It's a total misrepresentation of your data.

You have to round your result to match the physical uncertainty you actually achieved in the real world.

Right.

But let's consider a tool where the marks aren't so perfectly close together.

Imagine reading an analog voltmeter.

Okay.

The scale has printed marks for 0, 1, 2, all the way up to 10 volts.

But the physical needle is pointing somewhere in the wide empty space between the 5 and the 6.

Right.

Because the spacing is larger, you don't just pick 5 or 6, you use a technique called interpolation.

Exactly.

You use your eye to visually divide that empty space into invisible fractions.

You might reasonably decide the needle is a little less than halfway between the 5 and the 6.

So your best estimate is 5 .3 volts.

Right.

And since you are visually estimating that tenth of a volt, a reasonable probable range might be 5 .2 to 5 .4 volts.

Yeah.

And different students might argue over a single tenth of a volt, but everyone would agree that your estimated uncertainty window accurately reflects the physical limitations of the tool.

Exactly.

So we've established how to read the physical limitations of a tool, like a ruler or a dial.

But what happens when the tool is perfectly precise?

Like a digital stopwatch measuring to the hundredth of a second.

Right.

And the blurriness actually comes from the messy, flawed human holding it.

Oh, this is good.

Let's consider a classic experiment.

Timing the period of a swinging pendulum.

The main source of uncertainty is not reading the digital display, right?

It is your own unknown, fluctuating human reaction time, impressing the start and stop buttons.

Right.

Sometimes your thumb twitches a split second early.

Sometimes you blink and press it a split second late.

You can't just look at the stopwatch and magically estimate your own neurological delay.

Yeah, that's impossible.

So what's the mathematical solution here?

The mathematical solution here is repetition.

If you only time the pendulum once and get, say, 2 .3 seconds, you have zero context for your uncertainty.

True.

But suppose you repeat the measurement four times.

Your sequence of timings gives you 2 .3, 2 .4, 2 .5, and 2 .4 seconds.

Now mathematics actually allows us to make realistic estimates based on the spread of your own data.

OK, so the first assumption is that the best estimate of the period is just the average of those values.

You average them out and get 2 .4 seconds.

Right.

And the second assumption is that the true period probably lies somewhere between your lowest recorded value and your highest recorded value.

Makes sense.

Your quickest trigger finger gave you 2 .3.

Your slowest gave you 2 .5.

So your best estimate is 2 .4 seconds, with a probable range of 2 .3 to 2 .5 seconds.

Exactly.

Now, in later chapters, the text will introduce statistical methods showing that taking multiple measurements actually gives you a slightly smaller, more refined uncertainty than just taking the raw range from lowest to highest.

Oh really?

Yeah.

But for introductory purposes, using the spread of your repeated measurements gives you a simple, objective conclusion that relies on the actual data rather than just your personal guesswork.

OK, let me test a hypothesis here then.

If my human reaction time is the problem and averaging smooths it out, then if I just measure the pendulum swing a hundred times, my average gets better and better and my uncertainty basically disappears, right?

Ah, that is the trap that catches almost every young scientist.

Wait, it doesn't work.

No.

Repeating measurements only help smooth out random uncertainties.

Those are the fluctuations that bounce back and forth, like your reaction time being a little fast, then a little slow.

But what if your digital stopwatch is actually broken and it's internally running 5 % too fast?

Oh.

If the internal clock is running fast, then every single measurement I take is going to be wrong in the exact same direction.

Yes.

If the clock is fast, no amount of repeating the measurement will reveal the deficiency.

Your spread from lowest to highest will look beautifully tight and precise.

Your data will look incredibly consistent.

But the entire data set will be shifted off reality.

Exactly.

This is the very definition of a systematic error.

Wow.

That means I could do everything right.

I could measure it a hundred times, calculate my averages perfectly, draw beautiful error bars and still be completely disconnected from reality.

Yes.

Math can't save you there because the math only knows the numbers the broken watch gave it.

So what do you do?

The only remedy for a systematic error is calibration.

You absolutely have to check your measuring device against a more reliable one.

So I have to calibrate my stopwatch against a known, highly accurate laboratory clock.

Right.

You have to zero your scales.

You have to account for the temperature of the room expanding your metal rulers.

If there's any doubt about the underlying reliability of your tools, they must be tested.

Systematic errors are basically the invisible ghosts haunting every laboratory experiment.

That is the perfect way to describe it.

Well, let's step back and look at the ground we've covered today.

We started by completely rewiring how we think about the word error.

Right.

We aren't hunting for mistakes.

We are mapping inescapable uncertainty.

And we explored the problem of definition, realizing that even with a laser, a wooden door doesn't have a single exact height because the physical universe is constantly shifting.

Then we saw how George and Martha demonstrated that an experiment's validity relies entirely on justified ranges of uncertainty.

Yeah, a best estimate by itself is meaningless.

If the error bars are so wide, they overlap with multiple possible truths.

Exactly.

We also learned how to find those ranges using common sense on mark scales, interpolating the empty spaces between lines.

And being highly suspicious of the eight decimal places on our pocket calculators that imply we are measuring bacteria with plastic rulers.

Yes, please remember that one.

And finally, we saw how averaging repeated measurements can help smooth out our own clumsy human reaction times.

Provided we stay forever vigilant against the invisible threat of systematic errors shifting our entire data set.

Right.

You know, if you take nothing else away from this deep dive, take this philosophical truth.

The simple act of measuring a doorway or a pencil or the swing of a pendulum forces us to confront the fact that absolute perfect exactness is an illusion.

Wow.

We can never know the true numerical value of anything in the physical world with infinite precision.

True science isn't about achieving perfect exactness.

It is about learning exactly how uncertain we are, improving it.

That is such a powerful perspective to take to your lab bench tomorrow.

You aren't failing when you find uncertainty.

You are doing actual science.

Well, on behalf of the last minute lecture team, thank you for joining us for this tutoring session.

Take a breath, trust your common sense, and good luck in the lab.

You've got this.