WEBVTT
Kind: captions
Language: en-US

00:00:01.380 --> 00:00:02.980
[Silence]

00:00:02.980 --> 00:00:05.080
Good morning, everyone.

00:00:05.080 --> 00:00:07.859
Thank you all for coming
today to today’s lecture.

00:00:07.859 --> 00:00:12.290
Before the seminar begins, just a couple
of quick announcements for next week.

00:00:12.290 --> 00:00:18.320
Next week, we’ve got Dave Wald
coming on Tuesday giving a seminar.

00:00:18.320 --> 00:00:21.790
This is an extra one. And then also
next Wednesday, we have something

00:00:21.790 --> 00:00:25.100
a little bit different again.
We’ve got an anthropologist coming.

00:00:25.100 --> 00:00:29.000
His name is A.J. Faas, and he’s
from San Jose State University.

00:00:29.000 --> 00:00:32.710
And he studies people in volcanic
eruptions as well as earthquakes.

00:00:32.710 --> 00:00:36.840
So I’m really excited about that talk.
It’s one of my – my social scientist

00:00:36.840 --> 00:00:41.950
buddies coming to talk to you guys.
So without further ado, Nick Beeler

00:00:41.950 --> 00:00:47.800
is going to give the
introduction to our speaker.

00:00:49.620 --> 00:00:53.480
- Thanks, Sara.
So I know many of you know Greg,

00:00:53.480 --> 00:00:58.640
but some of you don’t.
Greg, once upon a time, in a land

00:00:58.640 --> 00:01:02.940
far away, he was a Mendenhall
postdoc here in Menlo Park.

00:01:02.940 --> 00:01:08.500
This was about five years ago. He left.
And he comes from an interesting

00:01:08.500 --> 00:01:11.760
background. He's an engineer.
He was an undergraduate at Cornell.

00:01:11.760 --> 00:01:15.600
Then he went to Berkeley and
worked with Steve Glaser, who is

00:01:15.600 --> 00:01:19.290
a non-destructive testing kind of guy,
basically doing engineering stuff that’s,

00:01:19.290 --> 00:01:22.930
like, lab-scale seismology. And he got
interested in earthquakes somewhere

00:01:22.930 --> 00:01:26.630
along the line. I think Roland Burgmann
may have been involved in this.

00:01:26.630 --> 00:01:31.430
And he ended up coming to – applying
for a Mendenhall, which he got.

00:01:31.430 --> 00:01:36.140
And he basically changed the
perception here in Menlo Park about

00:01:36.140 --> 00:01:40.270
the usefulness of the labs at the time.
This is – you know, we still are sort of

00:01:40.270 --> 00:01:46.320
riding the Greg wave. So people really
thought that he was the greatest thing

00:01:46.320 --> 00:01:48.820
that ever happened to the
labs and made them relevant.

00:01:48.820 --> 00:01:52.200
And that actually was true.
He brought in instrumentation

00:01:52.210 --> 00:01:55.450
that we couldn’t afford from
the Mendenhall program.

00:01:55.450 --> 00:01:58.350
He brought in expertise
that we didn’t have.

00:01:58.350 --> 00:02:03.320
He sort of influenced myself and
Dave Lockner and Brian Kilgore

00:02:03.320 --> 00:02:08.360
and the way we kind of approached
doing laboratory work and also relating

00:02:08.360 --> 00:02:11.120
it more directly and
specifically to earthquakes.

00:02:11.120 --> 00:02:14.890
And so, unfortunately,
he eventually packed his bags

00:02:14.890 --> 00:02:19.530
and went back to Cornell.
And, using some of the knowledge

00:02:19.530 --> 00:02:24.810
that he accumulated along the way,
he built the world’s largest laboratory

00:02:24.810 --> 00:02:28.630
faulting apparatus. Which I think
it still may be the largest one.

00:02:28.630 --> 00:02:31.550
It’s absolutely huge.

00:02:31.550 --> 00:02:34.080
And size matters, people.

00:02:35.030 --> 00:02:38.640
So what he’s going to talk about today
is earthquake nucleation, seismic

00:02:38.650 --> 00:02:42.810
radiation, and termination of dynamic
rupture in a 3-meter fault experiment.

00:02:42.810 --> 00:02:47.300
And so not only is this machine
bigger than the USGS one –

00:02:47.300 --> 00:02:52.040
really, in terms of area, about double.
Greg was just telling me.

00:02:52.040 --> 00:02:58.820
It also behaves in a fundamentally
different manner, essentially by design,

00:02:58.820 --> 00:03:03.040
and it can produce something that
we can’t easily do – we can do,

00:03:03.040 --> 00:03:06.650
but we can’t do it so easily, and that is,
we sort of produce characteristic

00:03:06.650 --> 00:03:10.200
earthquakes. And he produces complex
sequences, so there’s all kinds of

00:03:10.200 --> 00:03:14.780
different things that you can do –
stress transfer, earthquake prediction.

00:03:14.780 --> 00:03:16.980
In a more complicated system,
that can be done.

00:03:16.980 --> 00:03:20.620
And so, without,
further ado, Greg.

00:03:24.500 --> 00:03:28.500
- Thank you very much, and thanks,
Nick, for the kind introduction.

00:03:29.300 --> 00:03:30.920
Somewhat true.

00:03:30.920 --> 00:03:34.700
Okay. So I’ll be talking
about these lab experiments.

00:03:34.709 --> 00:03:37.569
Before I do that, I want to just lay
a little bit of groundwork for some of

00:03:37.569 --> 00:03:41.220
the things I’ll be talking about.
A number – you know, a number of

00:03:41.220 --> 00:03:45.810
models consider earthquakes to be shear
cracks, and I’m going to be adopting

00:03:45.810 --> 00:03:50.960
that, at least for part of this work.
And what a crack does is it initiates

00:03:50.960 --> 00:03:55.080
at some point along a fault.
It accelerates. It propagates.

00:03:55.080 --> 00:04:00.260
As it does that, it radiates seismic
waves. And then, eventually, it stops.

00:04:00.260 --> 00:04:04.220
And earthquakes are instabilities
that form when the crust weakens.

00:04:04.230 --> 00:04:08.769
So, if we have a measurement
location at some point on the fault,

00:04:08.769 --> 00:04:15.860
and a expanding earthquake or shear
crack, as the rupture goes by the fault,

00:04:15.860 --> 00:04:19.319
you might see a stress change that looks
a little bit like this, where the stress

00:04:19.319 --> 00:04:24.569
was initially at some level tau-zero.
It increases up to its peak and up to

00:04:24.569 --> 00:04:29.559
its strength – a peak stress, tau-p,
and then drops to a residual strength.

00:04:29.559 --> 00:04:32.110
And this weakening here, if you
plotted it – instead of as a function

00:04:32.110 --> 00:04:36.150
of time, if you plot it as a function
of slip, this weakening rate is

00:04:36.150 --> 00:04:39.860
really important because it controls –
basically an earthquake is an instability,

00:04:39.860 --> 00:04:44.310
and it’s an instability because that
weakening rate is faster than

00:04:44.310 --> 00:04:48.580
the strain energy can be released
with continued fault slip.

00:04:49.509 --> 00:04:53.159
So the questions that
motivate this work are many.

00:04:53.159 --> 00:04:56.930
Nick alluded to a number of them.
One question is, how does this

00:04:56.930 --> 00:05:00.430
type of a crack, or how does
this instability initiate?

00:05:00.430 --> 00:05:04.669
We often see foreshocks or small
seismic events close in space and time

00:05:04.669 --> 00:05:07.099
to the eventual hypocenter
of a larger earthquake.

00:05:07.099 --> 00:05:11.189
What’s the role of those foreshocks?
And then, thinking about small

00:05:11.189 --> 00:05:15.430
and large earthquakes, do they
both initiate similarly? Or are there –

00:05:15.430 --> 00:05:18.059
is a small – is a large earthquake
just a small one that got away

00:05:18.059 --> 00:05:20.460
for some reason?
Or are there some fundamental

00:05:20.460 --> 00:05:23.379
differences between small
and large earthquakes?

00:05:23.379 --> 00:05:26.909
What makes an earthquake rupture stop?
If that’s really what controls the

00:05:26.909 --> 00:05:30.860
magnitude of an earthquake, what
are the physics that control that?

00:05:30.860 --> 00:05:33.810
And then how are seismic waves
radiated during this process?

00:05:33.810 --> 00:05:37.219
We also know that earthquakes
come both in fast and slow types,

00:05:37.219 --> 00:05:39.789
and there’s tremor that’s radiated,
so if we can better understand

00:05:39.789 --> 00:05:43.960
the mechanics of seismic wave
radiation, that would be very useful.

00:05:44.720 --> 00:05:46.940
Before I get too much further,
I want to acknowledge

00:05:46.949 --> 00:05:50.520
some of my collaborators
and funding sources.

00:05:50.520 --> 00:05:53.880
Much of this work is done in
collaboration with Professor Dave

00:05:53.880 --> 00:05:58.040
Kammer, who was my colleague at
Cornell. Now he’s at ETH in Zurich.

00:05:58.040 --> 00:06:03.200
And we co-advise Huey Ke,
who is here today also – so if you

00:06:03.210 --> 00:06:07.009
were interested in speaking with him.
The work on dynamic rupture

00:06:07.009 --> 00:06:10.909
termination and using linear elastic
fracture mechanics is all his.

00:06:10.909 --> 00:06:15.979
And then the work with seismic wave
radiation is with my student, Bill Wu.

00:06:15.979 --> 00:06:19.439
And he also helped
build this machine.

00:06:19.439 --> 00:06:22.659
So if we think about rock mechanics
experiments, they come in a variety

00:06:22.659 --> 00:06:27.289
of scales. Oftentimes, we’re looking
at hand samples or some

00:06:27.289 --> 00:06:31.349
millimeter-size samples.
And the main – or one of the main ways

00:06:31.349 --> 00:06:36.339
in which rock mechanics is useful
is to understand the behavior of rocks.

00:06:36.339 --> 00:06:41.050
Their strength at a variety of pressures
and temperatures and slip speeds

00:06:41.050 --> 00:06:44.689
and perhaps even slip accelerations.
And, in these types of experiments,

00:06:44.689 --> 00:06:48.050
the sample is treated as a point.
So we would like to know how –

00:06:48.050 --> 00:06:52.440
if that sample was one point on a fault,
how it would behave, how its strength

00:06:52.440 --> 00:06:55.849
evolves with slip and time
and acceleration and stuff.

00:06:55.849 --> 00:06:59.319
And then, once we understand how
it behaves – how one point behaves,

00:06:59.319 --> 00:07:02.259
we can use that to inform
a constitutive relationship

00:07:02.259 --> 00:07:04.710
that will be put into
a computer model,

00:07:04.710 --> 00:07:10.800
and then we can extend it to a
continuum in a computer simulation.

00:07:10.800 --> 00:07:15.319
Another approach, and the one that
I’ve sort of adopted, is to use

00:07:15.319 --> 00:07:20.490
a larger sample. In this case, the sample
better approximates the continuum.

00:07:20.490 --> 00:07:23.259
It’s not just one point.
Now, with a large sample,

00:07:23.259 --> 00:07:26.830
it’s not possible to achieve realistic
pressures and temperatures.

00:07:26.830 --> 00:07:31.229
We’re typically confined to
lower stresses – 1 to 25 MPa.

00:07:31.229 --> 00:07:36.249
Most of the work I’ll be talking about
today is in the 5- to 10-MPa range.

00:07:36.249 --> 00:07:39.050
But, because the sample is bigger,
it behaves more like a continuum,

00:07:39.050 --> 00:07:41.919
and you can start to study
interactions from one part of a fault

00:07:41.919 --> 00:07:45.710
to another, which is not possible
with a smaller-scale sample.

00:07:45.710 --> 00:07:49.820
It also allows more room for
instrumentation and facilitates

00:07:49.820 --> 00:07:53.459
laboratory seismology, where basically
you put some sensors on the sample

00:07:53.459 --> 00:07:58.689
and treat it as if it was in Earth, and you
can do seismology in a similar way

00:07:58.689 --> 00:08:01.899
to regular seismology, and that’s
something that I’ve developed.

00:08:01.899 --> 00:08:06.060
The other thing, and perhaps a key
point is, with a large sample – well,

00:08:06.060 --> 00:08:09.620
with a small sample, if you want to
measure how the rock behaves when it

00:08:09.620 --> 00:08:14.800
slips fast, you need to accelerate with
a fast motor or a fast-acting clutch

00:08:14.800 --> 00:08:18.459
or some other high-speed loading
mechanism. But in a large sample,

00:08:18.459 --> 00:08:23.580
the fault itself is loaded by –
it’s loaded by its own instability.

00:08:23.580 --> 00:08:26.449
One part of the fault will spontaneously
nucleate a dynamic rupture, and the

00:08:26.449 --> 00:08:29.500
other part of the fault will get
slammed by that nucleation,

00:08:29.500 --> 00:08:33.380
and that’s a little bit more realistic
of what happens in the Earth.

00:08:34.640 --> 00:08:39.680
So, going – looking now at
large-scale loading machines,

00:08:39.690 --> 00:08:43.810
of course, I was very influenced
by the 2-meter block here.

00:08:43.810 --> 00:08:47.030
This is Brian Kilgore for scale,
so it’s a 2-meter-long fault cut

00:08:47.030 --> 00:08:50.030
in a meter-and-a-half by
meter-and-a-half slab of granite.

00:08:50.030 --> 00:08:54.460
I took a trip to Japan.
There’s a group at NIED in Tsukuba

00:08:54.460 --> 00:08:57.020
where they have a meter-and-a-half
sample – a little bit different

00:08:57.020 --> 00:09:00.220
configuration. So what I did is,
I kind of took these two designs,

00:09:00.220 --> 00:09:02.380
merged them together,
and created a machine

00:09:02.380 --> 00:09:07.360
at Cornell that uses
some attributes of both.

00:09:07.360 --> 00:09:10.700
And this is what it looks like.
It’s a 3-meter-long slab of granite.

00:09:10.700 --> 00:09:14.160
It’s not quite as thick.
It’s 0.3 meters thick.

00:09:14.160 --> 00:09:18.220
Not quite as thick as the USGS machine.
It’s designed to go to higher

00:09:18.220 --> 00:09:21.040
normal stress. The machine is
designed to go up to 20 MPa.

00:09:21.040 --> 00:09:24.361
We’ve only tested up to 12.
We’re going to try and get

00:09:24.361 --> 00:09:28.020
a few papers out before
we break the rock accidentally.

00:09:30.200 --> 00:09:33.460
By building this machine, I started to
develop a greater appreciation for what

00:09:33.460 --> 00:09:36.090
Jim Dieterich and other people
in the late ’70s went through

00:09:36.090 --> 00:09:40.680
to build the machine here.
I wanted to make a 4-meter block

00:09:40.680 --> 00:09:45.250
or a 10-meter block, but basically
I was limited by the size of the steel

00:09:45.250 --> 00:09:48.420
that could be shipped down
the road on a wide-load truck.

00:09:48.420 --> 00:09:53.140
So just unloading the steel
had a lot of engineering challenges

00:09:53.140 --> 00:09:56.420
associated with it.
Got it in the lab.

00:09:57.080 --> 00:10:00.520
Did a lot of grinding
and stacking.

00:10:00.520 --> 00:10:04.080
And then I had
a lot of self-doubt.

00:10:04.080 --> 00:10:06.600
Why am I doing this again?
[laughs]

00:10:06.600 --> 00:10:09.620
Why didn’t I just stay at the USGS?
They have a working machine.

00:10:09.630 --> 00:10:13.270
Anyway. [laughter]
We learned how to drill holes in steel

00:10:13.270 --> 00:10:16.690
and plumb high-pressure hydraulics,
and I had some help from my student,

00:10:16.690 --> 00:10:20.270
Bill Wu, and also Anthony Reid,
who is an undergraduate at Syracuse

00:10:20.270 --> 00:10:25.110
University. And, over the course of the
summer in 2016, we put this machine

00:10:25.110 --> 00:10:29.090
together. This is what the samples look
like when they’re not in the machine.

00:10:29.090 --> 00:10:32.440
And pretty much everything
had to be lifted with a crane.

00:10:32.440 --> 00:10:36.270
Even the smallest piece could,
you know, really hurt someone

00:10:36.270 --> 00:10:37.940
if it dropped on them.

00:10:37.940 --> 00:10:41.680
So basically, the way – this is
what the machine looks like today.

00:10:41.680 --> 00:10:47.140
The way that the – that it works is
there’s an array of hydraulic cylinders –

00:10:47.140 --> 00:10:51.320
these are the yellow things here –
that expand, and they squeeze these

00:10:51.330 --> 00:10:53.840
two blocks together.
So we’ve got two blocks – a moving

00:10:53.840 --> 00:10:57.460
block and a stationary block.
They squeeze the two blocks together.

00:10:57.460 --> 00:11:01.200
And then we have another array of
cylinders on the left side here that

00:11:01.200 --> 00:11:04.900
shears one block past each other.
And there’s a low-friction interface

00:11:04.900 --> 00:11:10.780
located behind this array of cylinders so
that basically the sample can translate.

00:11:10.780 --> 00:11:13.900
So we can have slip
on a rock/rock interface here,

00:11:13.900 --> 00:11:18.020
and then also slip on
this Teflon/steel interface.

00:11:18.020 --> 00:11:21.120
And the sample can translate
up to 50 millimeters before we

00:11:21.120 --> 00:11:24.020
have to lift it back up
with the crane and reset it.

00:11:25.820 --> 00:11:28.560
We have a lot of instrumentation
on the – on the block.

00:11:28.560 --> 00:11:32.140
We’ve got an
array of slip sensors.

00:11:33.340 --> 00:11:36.360
Shown by these little squares here.
I color-code them blue on one end,

00:11:36.360 --> 00:11:39.540
red on the other end. And so we
can measure the local fault slip

00:11:39.540 --> 00:11:42.770
at 16 locations. We’ve got an
array of piezoelectric sensors.

00:11:42.770 --> 00:11:44.810
They’re basically
single-component seismometers.

00:11:44.810 --> 00:11:48.380
We’ve got those on the top and bottom
of the sample to measure the radiated

00:11:48.380 --> 00:11:52.740
waves when the thing does rupture.
And we’ve got an array of strain gauge

00:11:52.740 --> 00:11:58.030
pairs to measure the local shear stress
along the fault. In addition to knowing

00:11:58.030 --> 00:12:03.000
the overall force on the sample in the
normal stress and shear stress directions.

00:12:04.270 --> 00:12:07.140
And hopefully there will be a lot of
exciting things to come out of this,

00:12:07.150 --> 00:12:10.010
but what’s – what came up initially,
and what we’re excited about now

00:12:10.010 --> 00:12:12.800
is that we’re able to
get these confined ruptures.

00:12:12.800 --> 00:12:18.700
Where part of the fault is able to slip,
and the rest of the fault doesn’t slip.

00:12:19.860 --> 00:12:22.840
This was possible on the 2-meter
block, but only with fluid injection.

00:12:22.840 --> 00:12:25.370
And we’re able to get
sequences of these events.

00:12:25.370 --> 00:12:28.540
And to show you an example
of what one of these looks like,

00:12:28.540 --> 00:12:32.390
if I color-code the slip sensors,
so red on one end, blue on the other end,

00:12:32.390 --> 00:12:35.720
and we plot them as a function of time,
a slip event is shown here.

00:12:35.720 --> 00:12:40.100
It’s a sudden increase in slip
on a number of slip sensors.

00:12:40.100 --> 00:12:43.620
But you can see that the blue
channel here didn’t slip at all.

00:12:43.620 --> 00:12:50.670
So this is a slip event that was confined
to this north end of the fault over here.

00:12:50.670 --> 00:12:54.290
And you can also see that there was
a little bit of slow slip that happened

00:12:54.290 --> 00:13:01.280
on a localized part of the fault before
this much faster rupture occurred.

00:13:01.280 --> 00:13:04.910
Now, I’m going to show – this plot
on the right here, I call it a pink-purple

00:13:04.910 --> 00:13:07.880
diagram. It’s a different way
of viewing the exact same data.

00:13:07.880 --> 00:13:12.560
Basically, instead of plotting the slip in
time, we’re plotting the slip in space.

00:13:12.560 --> 00:13:18.380
So, at any given moment,
we’ll make a little snapshot of slip.

00:13:18.380 --> 00:13:23.550
And that gives you the slip that was
measured at this array of 16 locations

00:13:23.550 --> 00:13:28.400
at a given time. We lay down
another snapshot every millisecond.

00:13:28.400 --> 00:13:34.000
So when those contours are right next
to each other, it’s slipping very slowly.

00:13:34.000 --> 00:13:37.090
When they separate, it’s slipping fast.
And I’ve changed the color of the

00:13:37.090 --> 00:13:41.360
contours from light pink to dark purple
so that, when you see this color banding,

00:13:41.360 --> 00:13:44.340
that means that
there’s slow creep.

00:13:44.340 --> 00:13:47.050
In this case, it’s about
20 microns per second.

00:13:47.050 --> 00:13:49.690
And then, when the contours start
to separate out, that creep is

00:13:49.690 --> 00:13:54.970
accelerating and then going dynamic.
So here you can see a nucleation region

00:13:54.970 --> 00:13:58.400
where it was slipping slowly
over about a meter of length.

00:13:58.400 --> 00:14:02.570
And then it started to expand out the
edge of that – out of the edge of that

00:14:02.570 --> 00:14:05.490
nucleation region and went dynamic.
Slipped very fast and then

00:14:05.490 --> 00:14:08.730
stopped quite abruptly.
And then this line here is kind of

00:14:08.730 --> 00:14:13.730
the overall distribution
of slip after it came to a halt.

00:14:13.730 --> 00:14:16.910
I’ll be showing you guys
a lot of pink-purple diagrams,

00:14:16.910 --> 00:14:19.450
so we’ll come back to that.
And one of the things that

00:14:19.450 --> 00:14:23.640
I’ll talk about a lot is this length
scale associated with this nucleation.

00:14:23.640 --> 00:14:26.700
And I’m going to call this h-star.
In this case, it’s about a meter.

00:14:26.700 --> 00:14:30.400
So, in other words, it takes – there’s a
slow-slipping patch – it’s about a meter

00:14:30.400 --> 00:14:35.980
in scale – that’s slipping before a
dynamic rupture event initiates.

00:14:36.980 --> 00:14:40.660
So the first question that we might
ask is, what causes a dynamic rupture

00:14:40.660 --> 00:14:45.400
to stop? What causes it to stop about
two-thirds of the way down this sample?

00:14:45.400 --> 00:14:50.360
And, to solve this, we’re going to rely
on linear elastic fracture mechanics.

00:14:50.360 --> 00:14:53.290
So going back to this plot,
where stress starts at some level,

00:14:53.290 --> 00:14:56.170
goes up to a peak strength,
and then drops to a residual strength,

00:14:56.170 --> 00:15:00.480
we can define a potential
stress drop, delta-tau-pot,

00:15:00.480 --> 00:15:06.990
which is the initial stress state,
minus the residual stress.

00:15:06.990 --> 00:15:11.130
But we can imagine that the
stresses along the fault vary, okay?

00:15:11.130 --> 00:15:15.220
So if we think of the peak stress as
this red line and the residual stress

00:15:15.220 --> 00:15:22.480
as this blue line, perhaps those
stresses vary along the fault.

00:15:22.480 --> 00:15:26.710
And then, the initial stress, the stress
state is this black dashed line.

00:15:26.710 --> 00:15:30.940
So we can define some sections of the 
fault where the stress is at a level that’s

00:15:30.940 --> 00:15:34.710
above the residual strength, meaning
that, if the fault slipped, it could – it could

00:15:34.710 --> 00:15:39.290
weaken and release some of its strain
energy, and that’s shown here in green.

00:15:39.290 --> 00:15:42.240
And then there’s other parts where
maybe the initial stress is so low that it’s

00:15:42.240 --> 00:15:44.710
below the residual
strength of the rock.

00:15:44.710 --> 00:15:49.050
And so, even if it slipped, it wouldn’t –
it wouldn’t weaken any further.

00:15:49.050 --> 00:15:53.450
And we can re-normalize this a little bit
so that, on the Y axis is just the potential

00:15:53.450 --> 00:15:59.100
stress drop, and find that there’s this
one region with positive potential stress

00:15:59.100 --> 00:16:01.900
drop, and another region with
negative potential stress drop.

00:16:01.900 --> 00:16:08.460
And, when we think about the way
in which a dynamic rupture or a shear

00:16:08.460 --> 00:16:12.300
crack would operate, we can think of it
as a competition between a strain

00:16:12.300 --> 00:16:16.589
energy release rate and a fracture energy.
The strain energy release rate

00:16:16.589 --> 00:16:19.589
is known to be a function
of this potential stress drop.

00:16:19.589 --> 00:16:24.430
So, if we could measure this,
we could calculate, using linear elastic

00:16:24.430 --> 00:16:29.000
fracture mechanics, the strain energy
release rate as a function of the location

00:16:29.000 --> 00:16:32.740
along the fault, if we assume
where this event is going to begin.

00:16:32.740 --> 00:16:37.860
And where that strain energy release
rate drops down below the fracture

00:16:37.860 --> 00:16:41.460
energy is theoretically where
that rupture should terminate.

00:16:42.240 --> 00:16:46.550
And what Huey Ke did in his
GRL paper from last year is use

00:16:46.550 --> 00:16:50.480
this to constrain, roughly, what
the fracture energy is on the fault.

00:16:50.480 --> 00:16:53.700
Because we can measure
the stress state before the rupture.

00:16:53.700 --> 00:16:56.650
We can measure it after the rupture.
And we also know where that rupture

00:16:56.650 --> 00:17:00.710
did terminate, and compare the model
prediction with where we observe it to

00:17:00.710 --> 00:17:05.939
terminate and get an estimate of the –
of the fracture energy that way.

00:17:05.939 --> 00:17:09.970
But the point I want you guys to come
away with at this point is that we’re

00:17:09.970 --> 00:17:14.709
able to manipulate the stress state on
the fault so that there is one rupture –

00:17:14.709 --> 00:17:19.679
one patch, of length p, that has
conditions favorable for rupture.

00:17:19.679 --> 00:17:24.730
And then, outside of that patch,
the rupture doesn’t want to propagate.

00:17:24.730 --> 00:17:29.420
The stress is so low that the rupture
will die. It’ll basically run out of gas.

00:17:29.420 --> 00:17:34.920
And, in order to get a laboratory
experiment where we get these confined

00:17:34.929 --> 00:17:37.960
ruptures, you need to have this patch
with favorable rupture conditions

00:17:37.960 --> 00:17:40.190
larger than a
nucleation length.

00:17:40.190 --> 00:17:43.389
Nucleation length is about a meter
for the stress levels that we work with.

00:17:43.389 --> 00:17:48.179
So that’s why we need a large block.
And we need to have that patch with

00:17:48.179 --> 00:17:51.639
favorable rupture conditions be smaller
than the overall sample length

00:17:51.640 --> 00:17:55.740
so that it will stop before it
reaches the ends of the sample.

00:17:57.960 --> 00:18:02.720
So let’s go through some typical
experiments on this machine.

00:18:02.730 --> 00:18:05.380
What we’ll typically do is we’ll
increase the normal stress up to

00:18:05.380 --> 00:18:09.760
some desired level, and then we’ll
hold it constant just by closing a valve.

00:18:09.760 --> 00:18:13.340
And then we’ll slowly increase
the shear stress on this sample.

00:18:13.340 --> 00:18:18.520
And we’ll eventually develop a
sequence of complete rupture events

00:18:18.529 --> 00:18:22.990
that are numbered, and they’re
characterized by this sudden drop

00:18:22.990 --> 00:18:25.999
in sample average shear stress.
And here I’m showing the slip

00:18:25.999 --> 00:18:27.740
just from two
different locations.

00:18:27.740 --> 00:18:32.240
And you can see that, whenever the
stress drops, the block slips ahead.

00:18:32.940 --> 00:18:36.540
But perhaps the most interesting
thing here is that, before we get

00:18:36.549 --> 00:18:39.659
these complete rupture events,
we get a sequence of events that

00:18:39.659 --> 00:18:46.250
don’t completely rupture the sample.
This is what the raw slip sensor data

00:18:46.250 --> 00:18:50.309
looks like. And if we plot them as
pink-purple diagrams, we can look at

00:18:50.309 --> 00:18:56.160
the Events 1 through 6 here and see that
Event #1 only slipped on a section of the

00:18:56.160 --> 00:19:01.179
fault maybe two-thirds of the way down.
Event #2 ruptured a little bit larger area.

00:19:01.179 --> 00:19:02.960
It also slipped a little bit faster.

00:19:02.960 --> 00:19:06.610
You can see these contours starting
to separate. Three is pretty similar.

00:19:06.610 --> 00:19:10.000
Four slips even faster.
Five is similar to 4.

00:19:10.000 --> 00:19:15.730
And then, by Event #6, it just
ruptures the entire interface.

00:19:15.730 --> 00:19:18.470
And you can also see here there’s –
you can see there’s sort of

00:19:18.470 --> 00:19:21.620
a nucleation process happening here.
This time, it’s nucleating about

00:19:21.620 --> 00:19:26.250
two-thirds of the way down the length
of the fault. And then it accelerates.

00:19:26.250 --> 00:19:31.360
And then, in this case, you can also –
in #4, you can see it ruptured,

00:19:31.360 --> 00:19:34.369
it kind of slowed down here, and then
there was sort of another episode of

00:19:34.369 --> 00:19:39.360
afterslip or something that occurred
on the – on the right side of the fault.

00:19:40.880 --> 00:19:45.480
So, to summarize these observations,
we can generate sequences of events.

00:19:45.480 --> 00:19:51.460
Some of them are slow – maybe peak
slip velocity of 100 microns per second.

00:19:51.460 --> 00:19:54.260
Some of them are fast – greater
than 10 millimeters per second.

00:19:54.260 --> 00:19:57.039
In this case, they’re all
nucleating at the same location,

00:19:57.040 --> 00:20:01.020
and successive events are
becoming larger and faster.

00:20:02.130 --> 00:20:04.700
So in order to understand the
mechanics of how this works,

00:20:04.700 --> 00:20:10.480
it’s important to understand how stress
changes during a confined slip event.

00:20:10.480 --> 00:20:14.409
So here I’m showing three different
examples of the distribution of slip

00:20:14.409 --> 00:20:19.309
during one of these slip events and the
distribution of shear stress changes.

00:20:19.309 --> 00:20:21.179
And you can see that,
where it slipped,

00:20:21.179 --> 00:20:25.090
or where the fault ruptured,
the stress dropped.

00:20:25.090 --> 00:20:27.649
And where – you know, right at
the edges of that ruptured region,

00:20:27.649 --> 00:20:31.340
the stress actually increased.
So the stress was kind of redistributed.

00:20:31.340 --> 00:20:34.820
It drops where it ruptures. It increases
near the edge of that rupture.

00:20:34.820 --> 00:20:38.240
Here’s a larger event that partially
ruptured through the end of the sample.

00:20:38.240 --> 00:20:42.820
You can see the stress dropped in
the middle of that ruptured region.

00:20:42.820 --> 00:20:45.580
But then it increases
near the edge of it.

00:20:46.360 --> 00:20:48.200
Same here.

00:20:48.200 --> 00:20:52.190
So, if we imagine that there was
some initial stress distribution

00:20:52.190 --> 00:20:57.269
along this fault shown with the
black line, then once that stress level

00:20:57.269 --> 00:21:01.830
reached a peak strength, and a slip event
occurred, the stress would drop

00:21:01.830 --> 00:21:05.450
within the ruptured region, and it would
increase on the surrounding regions

00:21:05.450 --> 00:21:08.700
so that the stress distribution
after the slip event is shown here

00:21:08.700 --> 00:21:11.110
in this light blue.
So what you can see

00:21:11.110 --> 00:21:15.960
is that it kind of smoothed out
this peak in the stress distribution.

00:21:16.760 --> 00:21:19.739
And then – so now here’s
our new stress distribution.

00:21:19.739 --> 00:21:23.149
We continue to load the sample.
It increases the shear stress.

00:21:23.149 --> 00:21:26.690
It reaches a critical stress level again,
and another slip event occurs.

00:21:26.690 --> 00:21:32.379
But now the section of the
fault that is – has a positive

00:21:32.379 --> 00:21:36.619
potential stress drop or stress level
above the residual stress is larger,

00:21:36.619 --> 00:21:39.169
and you get a
somewhat larger event.

00:21:39.169 --> 00:21:42.560
And then, again, the stress changes.
The stress drops within the ruptured

00:21:42.560 --> 00:21:46.360
region, increases a little bit at the edges
of the ruptured region, and we can

00:21:46.370 --> 00:21:50.639
get these sequences of events that
start out small, and then they grow

00:21:50.639 --> 00:21:54.230
with successive events and
become progressively faster.

00:21:54.230 --> 00:21:56.620
The reason that they’re getting
progressively faster is because

00:21:56.620 --> 00:21:59.850
this nucleation length scale
isn’t changing very much.

00:21:59.850 --> 00:22:04.110
But the patch, p, with conditions
favorable for a rupture, is getting

00:22:04.110 --> 00:22:08.200
larger and larger. So the earthquake
has more room to accelerate

00:22:08.200 --> 00:22:12.340
and get faster and also radiate
seismic waves more efficiently

00:22:12.340 --> 00:22:16.280
before it reaches unfavorable
stress conditions and dies.

00:22:19.760 --> 00:22:24.640
So I’m going to shift gears a little bit
and talk about earthquake initiation.

00:22:25.620 --> 00:22:31.800
So, when Jim Dieterich built the – or,
had built the 2-meter block here, I think

00:22:31.809 --> 00:22:35.679
one of the reasons why he wanted to do
that is to study earthquake initiation.

00:22:35.679 --> 00:22:39.710
He wanted to create a sample with stress
conditions that were relatively uniform

00:22:39.710 --> 00:22:43.850
so that he could study nucleation
in a way that would be consistent

00:22:43.850 --> 00:22:49.980
with a numerical simulation.
My machine is – the sort of idea

00:22:49.980 --> 00:22:52.590
is to do something that’s a bit
different and make a really

00:22:52.590 --> 00:22:57.000
non-uniform stress distribution
and see what happens.

00:22:58.860 --> 00:23:03.480
So there’s this model of an earthquake
initiation that kind of looks like this.

00:23:03.480 --> 00:23:07.879
Basically, you get some slow slip on
one part of the fault, and then,

00:23:07.879 --> 00:23:12.110
when it starts to slip, the stresses
drop within the slipped region

00:23:12.110 --> 00:23:15.460
and sort of increase towards
the edge of that slipped region.

00:23:15.460 --> 00:23:18.470
And because of that stress concentration
at the edges of the slipped region,

00:23:18.470 --> 00:23:23.039
the slipping region may want to
expand a little bit as it continues to slip.

00:23:23.039 --> 00:23:27.960
Once it reaches a critical length scale,
shown here as h-star, it goes unstable,

00:23:27.960 --> 00:23:33.980
and it accelerates very quickly and starts
propagating at a speed close to the

00:23:33.980 --> 00:23:39.320
shear wave velocity and the solid
and radiating seismic waves.

00:23:39.320 --> 00:23:42.919
So this is what we saw before.
So one question you might ask is,

00:23:42.920 --> 00:23:46.960
what is this critical length
scale for natural faults?

00:23:48.760 --> 00:23:52.759
So I’m showing a schematic of a fault,
and we might have some fault sections

00:23:52.759 --> 00:23:57.440
that are primarily locked, and then
other faults sections that are sort of

00:23:57.440 --> 00:24:01.190
transitional and show some
extended creeping regions.

00:24:01.190 --> 00:24:04.789
For the locked sections, it might be
reasonable to expect that this h-star

00:24:04.789 --> 00:24:07.970
is not that different from what we
see in the lab, that the nucleation

00:24:07.970 --> 00:24:10.019
length scale could be
something close to a meter.

00:24:10.019 --> 00:24:14.999
For creeping faults, h-star
could be much, much larger –

00:24:14.999 --> 00:24:20.340
perhaps tens of kilometers.
If you have a velocity-strengthening

00:24:20.340 --> 00:24:24.320
fault, then theoretically,
this h-star is infinity

00:24:24.320 --> 00:24:27.020
because an earthquake
can’t nucleate.

00:24:27.560 --> 00:24:32.400
And the nucleation length scale also
has important implications for the

00:24:32.409 --> 00:24:36.070
way we would interpret foreshocks.
If you have a large nucleation

00:24:36.070 --> 00:24:39.590
length scale – something on the order of
kilometers, then if you had some

00:24:39.590 --> 00:24:43.470
tiny events that rupture an area
that’s much less than kilometers,

00:24:43.470 --> 00:24:48.200
then those must be due to some
heterogeneous friction properties

00:24:48.200 --> 00:24:51.080
or heterogeneous strength.
And these are basically

00:24:51.080 --> 00:24:56.820
sub-h-star events, and they’re a
byproduct of some nucleation process.

00:24:56.820 --> 00:25:01.289
So if you get a little swarm of small
earthquakes, and you expect the h-star

00:25:01.289 --> 00:25:04.700
is very large, then that swarm is just
a byproduct of the nucleation

00:25:04.700 --> 00:25:07.540
of a larger earthquake, and those
small earthquakes are, in some way,

00:25:07.540 --> 00:25:10.060
fundamentally different
than the large earthquakes.

00:25:10.070 --> 00:25:13.960
Whereas, if you expect h-star to
be very small, then there’s –

00:25:13.960 --> 00:25:19.030
it’s not really creeping because
the slow – any sort of aseismic slip

00:25:19.030 --> 00:25:22.220
will soon transition
to a dynamic rupture.

00:25:22.220 --> 00:25:25.649
And all these little foreshocks that
might occur are just normal earthquakes

00:25:25.649 --> 00:25:30.809
that met unfavorable stress conditions
and sort of died prematurely.

00:25:30.809 --> 00:25:34.640
And any foreshocks would just
be directly triggering each other.

00:25:37.289 --> 00:25:40.280
So here’s some different examples
of how we’re able to image the

00:25:40.289 --> 00:25:43.620
initiation of dynamic rupture on
this 3-meter block at Cornell.

00:25:43.620 --> 00:25:46.799
Here’s a example of
a event at 10 megapascals.

00:25:46.799 --> 00:25:51.640
We can see there’s this localized
zone of slow and accelerating slip,

00:25:51.640 --> 00:25:54.940
and then it transitions
into dynamic rupture.

00:25:54.940 --> 00:25:57.559
In this case, h-star is
a little less than a meter.

00:25:57.559 --> 00:26:01.440
Here’s one of these confined
slip events at a lower stress level.

00:26:01.440 --> 00:26:04.629
We can still see that it’s nucleating
and accelerating, and then it

00:26:04.629 --> 00:26:08.809
doesn’t even rupture much of the fault.
And here’s an example of nucleation

00:26:08.809 --> 00:26:13.129
at the edge of a creeping region.
So the color banding here is showing

00:26:13.129 --> 00:26:16.940
that it was slowly creeping on
the first meter or so of the fault.

00:26:16.940 --> 00:26:19.850
And then, right at the edge of that
creeping region, we saw a dynamic

00:26:19.850 --> 00:26:24.539
event nucleate. And this is consistent
with some numerical models.

00:26:24.539 --> 00:26:29.989
Now, these nucleation images that
we can get are really a one-dimensional

00:26:29.989 --> 00:26:33.289
view of what’s at least
a two-dimensional process.

00:26:33.289 --> 00:26:36.870
So we’ve just got slip sensors on the top.
But, for some experiments,

00:26:36.870 --> 00:26:41.179
we put slip sensors both on the top
and on the bottom of the slab.

00:26:41.179 --> 00:26:44.350
And so we can compare the
nucleation that we would have imaged

00:26:44.350 --> 00:26:47.610
just using the top sensors versus
just using the bottom sensors.

00:26:47.610 --> 00:26:50.759
And on the right side here,
I’m roughly interpolating

00:26:50.759 --> 00:26:54.570
what the slip on the fault must have
been at various snapshots in time.

00:26:54.570 --> 00:26:59.749
And what we find is that it starts to
nucleate actually at the bottom of the –

00:26:59.749 --> 00:27:05.049
of the sample. And then that slow
slip migrates up towards the top.

00:27:05.049 --> 00:27:09.650
And then it switches, and it starts to
slip more at the top of the sample.

00:27:09.650 --> 00:27:14.850
And then, eventually, they sort of
work in unison, and rupture propagates

00:27:14.850 --> 00:27:20.399
along strike. So this is actually a very –
it’s a lot more complicated than

00:27:20.399 --> 00:27:24.279
what we might see. And so here you
can kind of see, it was – at the bottom,

00:27:24.279 --> 00:27:26.119
it was slipping very slowly.
And then, all the sudden,

00:27:26.119 --> 00:27:27.620
it started slipping really fast.

00:27:27.620 --> 00:27:31.220
And this is because of an
interaction with the free surface.

00:27:31.220 --> 00:27:35.369
So what we see as a sudden
acceleration of this nucleation process,

00:27:35.369 --> 00:27:37.770
when the slipping patch interacts
with the free surface, either at

00:27:37.770 --> 00:27:41.200
the top and bottom of the slab,
or at the end of the slab,

00:27:41.200 --> 00:27:46.740
just highlights the importance
of heterogeneity in this case.

00:27:46.749 --> 00:27:49.649
But fortunately, we can identify
this nucleation length scale.

00:27:49.649 --> 00:27:53.149
In this case, it’s a little less than a meter. 
We can identify it pretty confidently

00:27:53.149 --> 00:27:55.789
using either the top
or bottom sensors.

00:27:55.789 --> 00:27:59.560
So we can move forward and
look at the nucleation length scale.

00:28:00.390 --> 00:28:03.520
So what can we learn
from this nucleation?

00:28:03.539 --> 00:28:10.159
Well, first of all, we can compare a slip
event – this is Event #5 and #6 from

00:28:10.160 --> 00:28:15.000
the previous sequence I showed.
They both nucleated pretty similarly.

00:28:15.000 --> 00:28:19.900
And Event #5 met unfavorable stress
conditions and stopped, and Event #6

00:28:19.909 --> 00:28:24.409
ruptured through the entire interface.
And so, basically, what this shows

00:28:24.409 --> 00:28:28.809
is that stress conditions just a few
nucleation length scales away don’t have

00:28:28.809 --> 00:28:34.450
any effect, or have very minimal effect,
on the nucleation processes over there.

00:28:34.450 --> 00:28:37.240
So, in other words,
a nucleating earthquake

00:28:37.240 --> 00:28:40.080
doesn’t know how
big it will become.

00:28:41.060 --> 00:28:44.940
However, we do see a large variation
in the nucleation process.

00:28:44.940 --> 00:28:49.139
So if we look later on in the sequence
to Events 7 through 12 here, we can

00:28:49.139 --> 00:28:53.669
see that, typically, when we get
into this repetitive stick-slip cycle,

00:28:53.669 --> 00:28:57.280
they’re nucleating close to the end
where we’re pushing on the block.

00:28:57.820 --> 00:29:02.320
And Events #7, 8, 10, 11, and 12
all nucleate pretty similarly.

00:29:02.320 --> 00:29:05.140
For some reason, Event #9 nucleated
differently. It nucleated in

00:29:05.140 --> 00:29:09.180
a different spot. And the
nucleation was a little bit smaller.

00:29:09.180 --> 00:29:13.139
Now, Event #9 was the
biggest event of the sequence.

00:29:13.139 --> 00:29:17.979
So you can’t see it very well from here,
but Event #9 slipped about 190 microns

00:29:17.979 --> 00:29:22.179
on average. These other events
slipped 120 to 150 microns.

00:29:22.179 --> 00:29:25.269
And we know it’s not – it wasn’t
a bigger event just because it

00:29:25.269 --> 00:29:27.919
nucleated in a different spot.
Because Event #6 nucleated in

00:29:27.919 --> 00:29:31.700
that same spot, but it had a much
larger and slower nucleation process,

00:29:31.700 --> 00:29:35.139
and it was a much weaker event.
So basically, what this is saying is

00:29:35.139 --> 00:29:39.529
that a smaller nucleation, a smaller
and more abrupt nucleation, ends up

00:29:39.529 --> 00:29:44.680
leading to a more powerful rupture if
all other things are relatively constant.

00:29:47.420 --> 00:29:51.279
And then perhaps the most interesting
observation that I’ve come across so far

00:29:51.280 --> 00:29:57.920
is that this nucleation length scale can
vary considerably if the fault is kicked.

00:29:57.920 --> 00:30:01.580
And by kick, I mean it’s something
that causes this fault to go above

00:30:01.580 --> 00:30:06.710
steady state. We can think of a
stick-slip cycle or an earthquake cycle

00:30:06.710 --> 00:30:10.830
as being something that oscillates
around a steady state line.

00:30:10.830 --> 00:30:14.639
So you could have postseismic slip
and then interseismic healing,

00:30:14.639 --> 00:30:17.590
and then a nucleating – during the
nucleation phase of an earthquake,

00:30:17.590 --> 00:30:21.850
we’re above steady state. And then
we could get coseismic rupture.

00:30:21.850 --> 00:30:25.879
If you somehow kicked the fault above
steady state – far above steady state,

00:30:25.879 --> 00:30:30.210
either by suddenly increasing the
loading velocity or even by just letting

00:30:30.210 --> 00:30:34.239
that fault sit and heal for a
longer-than-normal time,

00:30:34.239 --> 00:30:38.929
and then resuming the loading
velocity at the same rate,

00:30:38.929 --> 00:30:41.970
that causes profound differences
in the way that the events nucleate.

00:30:41.970 --> 00:30:43.429
And I’m going to show
an example of this.

00:30:43.429 --> 00:30:45.900
But I’m not the only
one to see this effect.

00:30:45.900 --> 00:30:48.960
Actually, Kato in 1992
shows a similar effect.

00:30:48.970 --> 00:30:52.090
I’ve seen a similar effect on
a smaller sample, and then there’s

00:30:52.090 --> 00:30:55.360
a recent paper on plastic experiments
that show the same thing.

00:30:55.360 --> 00:30:58.750
And there’s also modeling
studies that support this.

00:30:58.750 --> 00:31:03.179
So here we’ve got a sequence
of events – 7, 8, 9 are going on here.

00:31:03.179 --> 00:31:07.129
Number 9 nucleates kind of normal.
And then what we do is we just

00:31:07.129 --> 00:31:10.889
pause the loading. We let it sit
for a minute, and then we

00:31:10.889 --> 00:31:14.139
continue to load at the same rate.
And Event #10 nucleates

00:31:14.139 --> 00:31:17.159
quite differently. It nucleates in
the same spot, but as you can see,

00:31:17.159 --> 00:31:19.710
it just sort of
initiates out of nothing.

00:31:19.710 --> 00:31:24.679
It takes less than a micron of slip for this
thing to accelerate to seismic speeds.

00:31:24.679 --> 00:31:27.249
And if you had to guess what the
nucleation length scale would be,

00:31:27.249 --> 00:31:29.159
it’s smaller than
the spacing of our sensors.

00:31:29.159 --> 00:31:33.340
It’s at least five times smaller
than these standard events.

00:31:33.340 --> 00:31:37.909
So possibly an order of magnitude
variation in the nucleation length scale.

00:31:37.909 --> 00:31:41.460
And then we can go on and see
that Event #11 was a partial rupture

00:31:41.460 --> 00:31:45.099
that caused Event #12
to nucleate at the other side.

00:31:45.100 --> 00:31:48.590
And then, by Event #13,
it kind of goes back to normal.

00:31:49.220 --> 00:31:56.660
But this decreased length scale
and more abrupt initiation,

00:31:56.669 --> 00:32:02.070
I think has some profound effects
on the way that earthquakes work.

00:32:02.070 --> 00:32:07.029
Before I get further into that,
I want to show some seismograms.

00:32:07.029 --> 00:32:11.760
We have earthquakes now – confined
events that are both slow and fast.

00:32:12.520 --> 00:32:16.940
And here I’m showing some
single-component seismograms.

00:32:16.940 --> 00:32:19.249
This is two different
stations in each case.

00:32:19.249 --> 00:32:22.320
And I’m showing four different events.
The blue one is slow. The green one

00:32:22.320 --> 00:32:25.799
is fast, and we’ve got
some intermediate events.

00:32:25.799 --> 00:32:29.700
If we look at the low-frequency signals
that are coming from these events,

00:32:29.700 --> 00:32:32.769
they’re all rupturing
basically the same patch of fault.

00:32:32.769 --> 00:32:34.590
And you can see that
they’re all very similar.

00:32:34.590 --> 00:32:38.560
It’s just that, as they’re getting
faster and faster, the ground motions

00:32:38.560 --> 00:32:41.220
are getting more and
more contracted in time.

00:32:41.220 --> 00:32:45.109
But if we look at the higher-frequency
signals – here’s the high-pass-filtered

00:32:45.109 --> 00:32:48.330
version – you can see that the fast
event's radiating high frequencies

00:32:48.330 --> 00:32:51.409
much more efficiently
than the slow events.

00:32:51.409 --> 00:32:54.429
Here’s the pink-purple
diagrams from these events.

00:32:54.429 --> 00:32:57.679
The first three are from the
sequence that I showed earlier.

00:32:57.679 --> 00:33:02.190
The fastest event is at a little bit higher
normal stress and after one of these

00:33:02.190 --> 00:33:05.139
holds where we’re able to shrink
the nucleation length scale and

00:33:05.140 --> 00:33:08.880
get this thing to be a
much more powerful rupture.

00:33:08.880 --> 00:33:12.940
And one thing to notice is that the
fast events slip more, even though

00:33:12.950 --> 00:33:17.029
the rupture area hasn’t changed much.
And basically that means that the static

00:33:17.029 --> 00:33:22.029
stress drop of these events is higher.
So the faster events are higher stress

00:33:22.029 --> 00:33:25.080
drop, and the slower
events are lower stress drop.

00:33:27.480 --> 00:33:33.620
And if we plot the maximum slip speed
of these events against their stress drop,

00:33:33.620 --> 00:33:38.240
we can see that the fastest events have
stress drops that are about 1/2 an MPa.

00:33:38.240 --> 00:33:41.359
The maximum slip speed is
about 100 millimeters per second.

00:33:41.359 --> 00:33:44.529
This is pretty close to what we would 
expect for a normal earthquake.

00:33:44.529 --> 00:33:48.659
It’s a little bit on the slow side.
It’s a little bit of a weak earthquake.

00:33:48.659 --> 00:33:53.970
And these fast earthquakes follow a
trend, and this is a linear relationship

00:33:53.970 --> 00:33:58.759
between slip velocity and stress drop,
and that’s what was predicted

00:33:58.759 --> 00:34:02.960
by Jim Brune in 1970.
And if we just extrapolate along

00:34:02.960 --> 00:34:07.820
this trend up to a 3 MPa stress drop,
we get meter-per-second slip speeds.

00:34:07.820 --> 00:34:11.390
This is a standard earthquake.
So I think that the fastest contained

00:34:11.390 --> 00:34:14.960
events that we’re able to generate in
the lab are a little bit on the weak side,

00:34:14.960 --> 00:34:18.280
but basically, they’re
standard earthquakes.

00:34:18.280 --> 00:34:22.159
However, when we get to slower
slip events, they don’t follow this

00:34:22.159 --> 00:34:25.120
relationship anymore,
and the slip speed can

00:34:25.120 --> 00:34:27.410
get down into the hundreds
of microns per second.

00:34:27.410 --> 00:34:32.120
The stress drop is very low –
on the order of tens of kilopascals.

00:34:33.730 --> 00:34:37.780
And then, in addition to looking at
the waveforms of these seismograms,

00:34:37.790 --> 00:34:41.500
we can also look at the source spectra.
Now, I’ve obtained these source spectra

00:34:41.500 --> 00:34:45.610
using an empirical Green’s function
approach using a ball impact as a

00:34:45.610 --> 00:34:48.860
empirical Green’s function source.
This is a technique that I developed

00:34:48.860 --> 00:34:51.690
when I was a Mendenhall postdoc here.
And so we can take a look at

00:34:51.690 --> 00:34:54.090
the source – this is the
displacement spectra and velocity

00:34:54.090 --> 00:34:58.680
spectra of these four events.
The green event – the fastest one –

00:34:58.680 --> 00:35:02.320
has a distinct corner frequency.
It has pretty much an omega-squared

00:35:02.320 --> 00:35:06.410
fall-off at high frequencies. And it
matches the Brune model pretty well.

00:35:06.410 --> 00:35:10.520
However, the slower events
become more and more depleted

00:35:10.520 --> 00:35:15.140
in energy near the corner frequency.
If you had to choose what the

00:35:15.140 --> 00:35:18.710
corner frequency is, it’s getting
to lower and lower frequency.

00:35:18.710 --> 00:35:22.180
And the – consequently,
the radiated energy

00:35:22.180 --> 00:35:24.880
is dropping by
orders of magnitude.

00:35:24.880 --> 00:35:29.110
So the same patch on the fault
on this dry piece of granite

00:35:29.110 --> 00:35:33.420
is rupturing in many different ways.
And just by small variations in the

00:35:33.420 --> 00:35:37.930
nucleation length scale, we can get
a totally fast normal earthquake

00:35:37.930 --> 00:35:40.460
or a super slow earthquake that
doesn’t really radiate

00:35:40.460 --> 00:35:43.580
any waves at all –
or at least hardly detectable.

00:35:44.740 --> 00:35:49.100
The other thing to point out is that,
riding on top of this

00:35:49.100 --> 00:35:54.700
is this really weak signal.
It’s kind of emergent and tremor-like.

00:35:55.400 --> 00:36:01.120
There’s also some similarities to
what we see with natural fast

00:36:01.130 --> 00:36:05.170
and slow earthquakes.
So Ide et al. showed that standard

00:36:05.170 --> 00:36:08.870
earthquakes should follow this scaling
relationship, whereas slow earthquakes

00:36:08.870 --> 00:36:12.210
seem to have this omega-squared falloff.
So if we compared two earthquakes

00:36:12.210 --> 00:36:14.980
of about the same magnitude,
a fast earthquake would have this

00:36:14.980 --> 00:36:17.520
pronounced corner frequency
and omega-squared fall-off

00:36:17.520 --> 00:36:21.290
above the corner frequency.
The slow earthquake might not have –

00:36:21.290 --> 00:36:25.350
might have a lower corner frequency
and an omega-to-the-negative-1 falloff.

00:36:25.350 --> 00:36:27.980
And that’s really similar
to what we’re seeing in the lab.

00:36:27.980 --> 00:36:31.670
So, though the mechanism by which
we’re generating slow earthquakes

00:36:31.670 --> 00:36:35.330
is probably different from the
mechanism that’s occurring in the Earth,

00:36:35.330 --> 00:36:40.240
the consequences in terms of seismic
waves seem to be very similar.

00:36:41.720 --> 00:36:46.730
The other thing is we’re also able to
see this sort of weak tremor-like signal

00:36:46.730 --> 00:36:49.190
superimposed on top of
a lower-frequency signal.

00:36:49.190 --> 00:36:52.200
And that’s been observed in a couple
cases in nature as well, where you have,

00:36:52.200 --> 00:36:55.880
like, a VLFE event, or a very low
frequency earthquake, and then,

00:36:55.880 --> 00:36:59.060
superimposed on top of that is
a tremor-like signal, suggesting

00:36:59.060 --> 00:37:03.200
that these are sort of different
ways of imaging the same process.

00:37:03.200 --> 00:37:06.520
It’s just a weakly
radiating earthquake.

00:37:07.660 --> 00:37:12.420
So, to wrap up this part of the talk,
we can – we can generate

00:37:12.430 --> 00:37:16.670
these contained slip events.
We think that the rupture stops

00:37:16.670 --> 00:37:20.530
basically because they run out of gas.
They reach stress conditions

00:37:20.530 --> 00:37:24.820
that are unfavorable.
The stress just isn’t high enough.

00:37:24.820 --> 00:37:28.400
And then we can find a full spectrum
of slow to fast events under

00:37:28.400 --> 00:37:34.750
essentially identical conditions.
The fast events are normal earthquakes.

00:37:34.750 --> 00:37:37.140
The Brune spectrum works pretty well.
They’ve got 1 kilohertz

00:37:37.140 --> 00:37:40.800
corner frequencies
They’re magnitude negative 2.5.

00:37:40.800 --> 00:37:44.550
The slow events are low-frequency
earthquakes, and they exhibit a number

00:37:44.550 --> 00:37:48.120
of characteristics that are similar
to natural low frequencies.

00:37:48.120 --> 00:37:51.530
They have low stress drops,
low slip speeds, and the spectra

00:37:51.530 --> 00:37:54.620
is depleted near
the corner frequency.

00:37:54.620 --> 00:37:59.420
And then these slow events are
occurring because the patch that

00:37:59.420 --> 00:38:03.130
they rupture with favorable rupture
conditions is just barely larger,

00:38:03.130 --> 00:38:06.660
or maybe even smaller, than the
critical nucleation length scale.

00:38:06.660 --> 00:38:11.390
So, as this instability is trying to
develop, it basically runs out of gas

00:38:11.390 --> 00:38:14.980
before it fully develops and
reaches seismic slip speeds.

00:38:15.920 --> 00:38:21.720
So this is just a diagram to show,
if you have a patch of the same size,

00:38:21.720 --> 00:38:25.360
if you had a smaller nucleation length
scale, there would be more space

00:38:25.360 --> 00:38:30.540
on the fault, more fuel for the fault
to accelerate, than otherwise.

00:38:34.280 --> 00:38:38.660
Now, in the Earth – in the lab,
we set up a patch just by squeezing

00:38:38.670 --> 00:38:41.920
the two blocks together.
This sets up this kind of strange stress

00:38:41.920 --> 00:38:48.620
condition where we have a high
stress-over-strength ratio, right,

00:38:48.620 --> 00:38:50.010
about two-thirds of the
way down the fault

00:38:50.010 --> 00:38:54.760
and low stress-over-strength
ratio near the ends of the sample.

00:38:55.180 --> 00:38:59.820
But, as a couple of events occur,
it erases this stress distribution.

00:38:59.820 --> 00:39:02.930
So this isn’t sustainable.
If we just continue to load at

00:39:02.930 --> 00:39:05.400
a constant rate, we’ll eventually
get complete ruptures that

00:39:05.400 --> 00:39:08.430
blow through the whole fault.
But, in the Earth, we get

00:39:08.430 --> 00:39:12.430
repeating earthquake sequences.
We get tremor that seems to persistently

00:39:12.430 --> 00:39:18.680
occur on tremor asperities or some sort
of thing that’s persistent geographically.

00:39:18.680 --> 00:39:22.670
So some other mechanism must be
setting up this favorable rupture patch.

00:39:22.670 --> 00:39:26.530
And some ideas that have been proposed
is to have a small velocity-weakening

00:39:26.530 --> 00:39:29.720
patch surrounded by
velocity-strengthening material.

00:39:29.720 --> 00:39:34.230
Or maybe there’s just some sort of
a bump on the fault that causes

00:39:34.230 --> 00:39:38.830
an unstable patch surrounded
by more stable material.

00:39:38.830 --> 00:39:42.680
But if you take this idea, and you
combine it with this kick dependency to

00:39:42.680 --> 00:39:48.610
nucleation length scale, than you can –
you can imagine a case where you’ve

00:39:48.610 --> 00:39:53.310
got a fault where you’ve got some of
these patches that are, in general,

00:39:53.310 --> 00:39:57.110
so small that their nucleation length
scale is larger than the patch,

00:39:57.110 --> 00:39:59.880
so they really wouldn’t
radiate any seismic waves.

00:39:59.880 --> 00:40:04.260
But if the fault gets kicked by
something – perhaps a slow-slip front

00:40:04.260 --> 00:40:09.300
associated with a ETS event in
Cascadia or a VLF event or some

00:40:09.300 --> 00:40:12.450
other perturbation on the fault,
then the nucleation length scale

00:40:12.450 --> 00:40:18.420
can effectively decrease.
An event can initiate more abruptly.

00:40:18.420 --> 00:40:22.170
And these patches, which would
normally just slip silently, can weakly

00:40:22.170 --> 00:40:28.250
radiate. And this might be one of the
mechanisms by which tremor occurs.

00:40:28.250 --> 00:40:30.300
It also can help explain
some of the variation

00:40:30.300 --> 00:40:33.680
in repeating earthquake sequences
that we sometimes observe.

00:40:33.680 --> 00:40:36.700
So basically, if you’ve got some
slow-slip front or something that

00:40:36.700 --> 00:40:41.980
propagates down a fault, that can rapidly
load certain sections of the fault and

00:40:41.980 --> 00:40:47.470
cause these kick-dependency changes
in nucleation characteristics and then

00:40:47.470 --> 00:40:52.900
cause the profound changes in seismic
wave radiation associated with that.

00:40:54.080 --> 00:40:57.620
It also has some
implications for foreshocks.

00:40:58.680 --> 00:41:02.110
If we think about this pre-slip model,
or cascade model, these are really

00:41:02.110 --> 00:41:06.250
end member models in terms
of what we would expect.

00:41:06.250 --> 00:41:09.720
And there’s lots of room for
models in between those two.

00:41:09.720 --> 00:41:12.790
So you can imagine a case where you
have a large nucleation length scale,

00:41:12.790 --> 00:41:17.610
and you’ve got some small events
that are byproducts of that nucleation.

00:41:17.610 --> 00:41:22.130
But every time a small event occurs,
it ruptures quite quickly.

00:41:22.130 --> 00:41:24.910
And that essentially kicks
the surrounding fault.

00:41:24.910 --> 00:41:28.490
So those small events are
kind of testing the waters.

00:41:28.490 --> 00:41:31.870
They’re probing the fault to see, are
you ready to rupture dynamically?

00:41:31.870 --> 00:41:34.960
Because, when it gets kicked,
it becomes much more unstable

00:41:34.960 --> 00:41:38.610
and prone to dynamic rupture.
So you can imagine a case where you

00:41:38.610 --> 00:41:42.200
have a large nucleation length scale.
Maybe, because of some heterogeneity,

00:41:42.200 --> 00:41:45.390
you can get smaller events.
When one those events occurs,

00:41:45.390 --> 00:41:50.400
it can cascade up and immediately
transition into a large dynamic rupture.

00:41:50.400 --> 00:41:54.000
And that could – so basically,
it can jump-start dynamic rupture.

00:41:54.000 --> 00:41:56.410
And that can help resolve
some observations that small

00:41:56.410 --> 00:41:59.530
and large earthquakes
appear to initiate similarly.

00:41:59.530 --> 00:42:03.900
Because, even though there might be
a large nucleation process, in this case,

00:42:03.900 --> 00:42:08.630
the large nucleation was just one of
these small earthquakes that got away.

00:42:11.540 --> 00:42:17.320
Okay, so, to conclude this part,
we’re able to image the initiation

00:42:17.320 --> 00:42:24.140
process on this 3-meter fault.
And we find that, even though

00:42:24.140 --> 00:42:27.800
1 meter is a pretty good rough guess,
there’s significant variation in the

00:42:27.800 --> 00:42:31.360
way that these events nucleate.
Sometimes a kick above steady state

00:42:31.360 --> 00:42:33.480
or something can cause
that nucleation to shrink.

00:42:33.480 --> 00:42:36.500
Heterogeneity also
has a strong influence.

00:42:37.480 --> 00:42:41.940
I can talk about that if there’s –
if there’s interest.

00:42:41.940 --> 00:42:44.130
And that, if you do have
a smaller nucleation,

00:42:44.130 --> 00:42:49.140
or a more abrupt nucleation,
that causes a more energetic rupture.

00:42:51.530 --> 00:42:54.240
And these kick-induced variations
in nucleation length scale may

00:42:54.240 --> 00:42:59.260
explain both the variation in repeating
earthquake sequences and also the way

00:42:59.260 --> 00:43:04.580
that low-frequency earthquakes are
generated that compose tectonic tremor.

00:43:04.580 --> 00:43:08.920
I think I’ll end there and open up for
questions. Thank you very much.

00:43:08.920 --> 00:43:12.140
And I’d love to
have a nice discussion.

00:43:12.140 --> 00:43:18.260
[Applause]

00:43:19.240 --> 00:43:22.420
- That was a great talk. Right.
I’m sure we have some questions.

00:43:22.420 --> 00:43:24.760
Who wants to start?

00:43:27.100 --> 00:43:32.900
[Silence]

00:43:33.640 --> 00:43:35.640
- [inaudible]

00:43:38.880 --> 00:43:42.740
- Greg, where does the
heterogeneity tend to come from?

00:43:42.740 --> 00:43:49.320
Is it the surface of the fault?
Is it variations in loading?

00:43:49.320 --> 00:43:51.270
- Yeah.
- Over the 3-meter block?

00:43:51.270 --> 00:43:54.940
- You’re talking about in the lab, right?
- Yeah.

00:43:55.460 --> 00:43:57.400
- Yeah. So this is what
the fault surface looks like.

00:43:57.400 --> 00:43:59.910
It’s – you know,
it’s pretty flat and smooth.

00:43:59.910 --> 00:44:03.500
We do get higher normal stress
at the ends of the fault

00:44:03.500 --> 00:44:08.320
because of an edge effect.
And that tends to cause nucleation

00:44:08.320 --> 00:44:12.100
to start closer to the center
rather than right at the end.

00:44:12.110 --> 00:44:16.960
But the main way that the –
that the nucleation – that we’re able

00:44:16.960 --> 00:44:21.350
to sort of control it the way that
we do is by – basically all we do

00:44:21.350 --> 00:44:25.150
is increase the normal stress.
So we’re able to generate these

00:44:25.150 --> 00:44:29.370
sequences right after a pretty
large increase in normal stress.

00:44:29.370 --> 00:44:33.560
So what happens is, the bottom block –
the stationary block is a bit more

00:44:33.560 --> 00:44:37.230
constrained than the top block.
So as we increase the normal stress,

00:44:37.230 --> 00:44:39.620
there’s a Poisson expansion
effect that happens.

00:44:39.620 --> 00:44:43.020
So the top block wants to expand.
And what that does is it increases

00:44:43.020 --> 00:44:46.330
the shear stress on the right side
of the fault and decreases the

00:44:46.330 --> 00:44:48.800
shear stress on the
left side of the fault.

00:44:48.800 --> 00:44:53.100
So it’s a specific loading procedure that
causes this heterogeneity in shear stress.

00:44:53.100 --> 00:44:57.460
So when you combine that variation
in shear stress with this bowl-shaped

00:44:57.460 --> 00:45:00.000
variation in normal stress,
we end up with kind of like

00:45:00.000 --> 00:45:03.130
a sweet spot for nucleation
about two-thirds of the way down.

00:45:03.130 --> 00:45:06.150
But, once we generate
one of these sequences,

00:45:06.150 --> 00:45:08.750
that heterogeneity
is erased, basically.

00:45:08.750 --> 00:45:13.480
And so we just get complete rupture
events. That’s a loading procedure.

00:45:16.660 --> 00:45:18.540
- Thank you very much.
It was a great talk.

00:45:18.540 --> 00:45:24.340
I just a have a question about the
roughness of the – of the interface.

00:45:24.350 --> 00:45:26.640
- Mm-hmm.
- And whether – and how you

00:45:26.640 --> 00:45:34.150
decided on what roughness
to have the – to make the slabs.

00:45:34.150 --> 00:45:37.410
And I wonder if it changes with use?
Thank you.

00:45:37.410 --> 00:45:39.320
- Yeah. Great questions.

00:45:39.320 --> 00:45:42.100
It absolutely does change.
I’m going to try and get back to

00:45:42.100 --> 00:45:46.180
that picture of the fault.
So basically, what I wanted to –

00:45:46.180 --> 00:45:49.330
what I wanted to do is
make a fault that was very flat.

00:45:49.330 --> 00:45:51.510
Because I wanted to make sure
that it was going to be in contact

00:45:51.510 --> 00:45:56.230
everywhere along its length.
I didn’t prescribe anything special

00:45:56.230 --> 00:46:00.050
about its roughness – just as
flat and smooth as possible.

00:46:00.050 --> 00:46:05.010
And then, as the fault continues to
wear, the roughness does change.

00:46:05.010 --> 00:46:08.940
So what we did initially is we did
a run-in period where we loaded

00:46:08.940 --> 00:46:12.570
up to 7 megapascals and just
forced it to slip many, many times.

00:46:12.570 --> 00:46:15.240
We saw it start to strengthen and
become much more unstable.

00:46:15.240 --> 00:46:18.740
And this is similar to what’s
observed on smaller-scale samples.

00:46:18.740 --> 00:46:22.240
However, over time, you can see that
there’s some wear that’s occurring at

00:46:22.240 --> 00:46:28.430
the edge, and there’s some little grooves
that are starting to form in some places.

00:46:28.430 --> 00:46:33.010
Higher density of grooves
towards the far end of the sample.

00:46:33.010 --> 00:46:39.590
And we also observed differences if we
reset the block, and we take this gouge

00:46:39.590 --> 00:46:43.740
layer that’s developing over time – we
usually don’t wipe off the gouge at all.

00:46:43.740 --> 00:46:46.860
If we reset the block,
it resets the gouge layer a little bit.

00:46:46.860 --> 00:46:49.190
And we get a little bit different
behavior. You know, we have to –

00:46:49.190 --> 00:46:53.690
we have to kind of run it in again in
order to get it to become more unstable.

00:46:53.690 --> 00:46:57.500
And then, in January, we opened up
the block and tried to map the grooves.

00:46:57.500 --> 00:46:59.980
And, after we had touched the
fault a lot, now we’re getting

00:46:59.980 --> 00:47:05.530
a lot more creep on the fault.
So we’re – basically, the methodology

00:47:05.530 --> 00:47:08.790
we’ve been trying to do is
just document everything.

00:47:08.790 --> 00:47:13.180
And see – do some experiments that
we can repeat from the early days of

00:47:13.180 --> 00:47:18.330
this experiment to the later days and
see how the subtle changes in friction

00:47:18.330 --> 00:47:23.620
properties are manifesting themselves
in changes in the rupture properties.

00:47:25.720 --> 00:47:27.880
- That was a great talk, Greg.
I had a question.

00:47:27.880 --> 00:47:33.980
You had a nice observation that
showed that, if you had a, if you will,

00:47:33.980 --> 00:47:37.610
a more energetic nucleation – a small
nucleation patch with a large stress

00:47:37.610 --> 00:47:43.380
drop, that you got a bigger rupture.
So that’s sort of contrary to – for hope

00:47:43.380 --> 00:47:47.260
that everyone who does early warning,
you know, it’s sort of the opposite case.

00:47:47.260 --> 00:47:52.710
You know, the nucleation patch
size scales inversely with the

00:47:52.710 --> 00:47:56.720
eventual magnitude size.
But at the same time, it’s a –

00:47:56.720 --> 00:48:01.430
there’s a moment release rate, or there’s
aspects of this that you can detect.

00:48:01.430 --> 00:48:07.340
And so you could, in principle –
or could you, in principle, predict,

00:48:07.340 --> 00:48:13.010
in real time, eventual rupture
sizes based on the characteristics

00:48:13.010 --> 00:48:17.460
of the early stage of nucleation
or the early stage of rupture?

00:48:18.100 --> 00:48:23.820
- Yeah, okay. So, first of all, yeah,
real time stuff is very challenging,

00:48:23.830 --> 00:48:27.430
as I’m sure everyone in early warning
knows. So we don’t usually do that.

00:48:27.430 --> 00:48:30.780
We have some data that’s being
displayed as we do the experiments,

00:48:30.780 --> 00:48:33.540
and so we sort of have some
intuition for what’s going to go on.

00:48:33.540 --> 00:48:37.690
But when we want to look at the
details, everything is done offline.

00:48:37.690 --> 00:48:39.680
So we haven’t invested
in the type of infrastructure

00:48:39.680 --> 00:48:43.020
to do this sort of
thing in real time.

00:48:44.780 --> 00:48:49.180
Yeah, so prediction – well, we know –
we can predict what’s going on because

00:48:49.190 --> 00:48:53.360
we’ve done the experiment before.
[laughter]

00:48:53.360 --> 00:48:55.840
We often – you know,
we do something different,

00:48:55.840 --> 00:48:58.780
and it behaves
a little bit differently.

00:48:58.780 --> 00:49:03.580
But, in terms of this small nucleation
length leading to a more powerful

00:49:03.580 --> 00:49:07.980
rupture, this is something that
probably we see an exaggerated

00:49:07.980 --> 00:49:11.640
view of that in the lab.
So if you have a rupture patch that’s

00:49:11.640 --> 00:49:15.790
only maybe three or five times as
large as the nucleation length scale,

00:49:15.790 --> 00:49:20.180
then small variations in that nucleation
process lead to pretty large variations

00:49:20.180 --> 00:49:23.580
in the amount of area
that’s available to rupture.

00:49:23.580 --> 00:49:26.720
But if you had a different case where
the nucleation length scale is just

00:49:26.720 --> 00:49:32.270
1/1,000th of the total rupture length,
which might be a reality for natural

00:49:32.270 --> 00:49:35.690
faults, then those variations
probably have nothing to do

00:49:35.690 --> 00:49:40.300
with the power of the –
of the eventual rupture.

00:49:40.300 --> 00:49:43.920
So I think we’re getting
an exaggerated view in the lab.

00:49:44.700 --> 00:49:48.840
But for something like a small
repeating earthquake sequence,

00:49:48.840 --> 00:49:51.780
maybe that’s not the case.
Maybe for a small repeater

00:49:51.780 --> 00:49:55.310
or a low-frequency earthquake
or something like that,

00:49:55.310 --> 00:49:59.260
you would see this large variation
with nucleation length.

00:50:01.520 --> 00:50:05.760
- Yeah. Very interesting, Greg.
I think you just answered the question

00:50:05.760 --> 00:50:10.550
I was going to ask, which is,
we’ve known for a long time

00:50:10.550 --> 00:50:17.680
that your h-star there, the rupture –
the critical patch size for rupture goes

00:50:17.680 --> 00:50:22.560
down with – or, yeah, it goes down
with increasing normal stress –

00:50:22.560 --> 00:50:30.020
the increasing strength of the fault.
So if you extrapolated your results

00:50:30.020 --> 00:50:37.260
to typical earthquake depths, where
rupture initiates at, say, 10 kilometers,

00:50:37.260 --> 00:50:45.920
the patch size would be very small.
And it seems like it’s difficult to

00:50:45.920 --> 00:50:53.520
avoid the reality that the rupture patch
at depth needed for rupture is very

00:50:53.520 --> 00:50:58.900
small indeed, which I guess is bad
news for earthquake early warning.

00:50:58.900 --> 00:51:00.440
- Well, certainly bad news

00:51:00.440 --> 00:51:06.240
for earthquake forecasting or prediction,
right? I think that that idea is right.

00:51:06.960 --> 00:51:12.660
If the only thing that changes is the
normal stress, you would expect this

00:51:12.660 --> 00:51:15.380
nucleation length to be even smaller
than what we would see in the lab

00:51:15.380 --> 00:51:20.740
because it’s an inverse proportionality
to a nucleation length scale and a –

00:51:20.740 --> 00:51:23.740
and normal stress.
So if you’re at 100 megapascals,

00:51:23.740 --> 00:51:28.340
you might get 1/10 of a meter or
even smaller nucleation length scales.

00:51:28.340 --> 00:51:32.560
However, other factors
could change considerably.

00:51:32.560 --> 00:51:40.380
If you have a large damage zone
or basically a large gouge thickness

00:51:40.380 --> 00:51:44.730
or something like that, that could
increase the nucleation length scale.

00:51:44.730 --> 00:51:48.980
And also clay minerals and other
minerals that are known to be

00:51:48.980 --> 00:51:53.880
velocity-strengthening would
increase the nucleation length scale.

00:51:53.890 --> 00:51:59.410
And I think that subduction zones
in particular and also creeping sections

00:51:59.410 --> 00:52:03.970
of some faults here in California,
that’s evidence that this nucleation

00:52:03.970 --> 00:52:09.070
length scale must be larger – something
on the order – if you can get a section

00:52:09.070 --> 00:52:14.590
of the fault to slip, as has been imaged
geodetically in Tohoku, for example –

00:52:14.590 --> 00:52:20.160
if you can get kilometer-size sections to
slip without initiating an earthquake,

00:52:20.160 --> 00:52:23.970
that means that, at least at that
section of the fault, this length scale,

00:52:23.970 --> 00:52:28.240
if that’s the appropriate
measure to use, is quite large.

00:52:29.250 --> 00:52:34.140
But I think – I think you’re right
for probably continental faults

00:52:34.140 --> 00:52:38.710
or faults where you would
expect similar conditions

00:52:38.710 --> 00:52:42.780
to what we have in the lab,
where it’s bare rock on rock.

00:52:44.820 --> 00:52:48.280
- Hey, Greg. Thanks.
Amazing talk.

00:52:48.280 --> 00:52:52.320
I want to ask maybe Nick’s
question a little differently.

00:52:52.320 --> 00:52:54.140
- Better.
[laughter]

00:52:54.140 --> 00:52:58.440
- Can you give us an idea of, like,
the statistics of the whole life of

00:52:58.450 --> 00:53:01.820
the machine? Like, how many
overall events have you run?

00:53:01.820 --> 00:53:03.270
How many individual
events do you get?

00:53:03.270 --> 00:53:07.960
And then, do you see – like,
how strong is the correlation between,

00:53:07.960 --> 00:53:12.260
say, the h-star and the
energetics of the earthquake?

00:53:12.260 --> 00:53:15.240
And you gave us sort of
some really nice examples,

00:53:15.240 --> 00:53:19.480
but how repeatable is that behavior?
And so, to get to Nick’s question,

00:53:19.490 --> 00:53:23.750
if you could quickly measure that h-star,
how indicative of the – is that of the

00:53:23.750 --> 00:53:29.520
final energetics of the earthquake?
And how reliable is that?

00:53:32.200 --> 00:53:36.400
- Okay. So two questions.
So the h-star and energetics.

00:53:36.400 --> 00:53:40.290
The more I think about this problem,
the more I think that a nucleation length

00:53:40.290 --> 00:53:44.100
scale is probably not the appropriate
way to characterize nucleation.

00:53:45.480 --> 00:53:49.050
This h-star parameter is probably –
it’s an important parameter to sort of

00:53:49.050 --> 00:53:56.270
characterize the overall behavior
of the fault, but I have a lot of

00:53:56.270 --> 00:53:59.210
evidence starting to pile up that
say maybe this length scale is

00:53:59.210 --> 00:54:01.760
not really the right
way of thinking about it.

00:54:03.110 --> 00:54:08.460
So basically, going back to this
kick-assisted cascade-up model,

00:54:08.460 --> 00:54:11.490
where, even here, when you had
a really large length scale,

00:54:11.490 --> 00:54:16.610
a small event could
kick-start a rupture.

00:54:16.610 --> 00:54:20.520
So didn’t matter that the length
scale was on the order of kilometers.

00:54:20.520 --> 00:54:26.120
A 10-meter-size patch
could go – could go dynamic.

00:54:27.040 --> 00:54:33.060
Because of stress concentrations,
2D effects, and things like that.

00:54:34.820 --> 00:54:36.860
So maybe that didn’t
answer it any better.

00:54:36.860 --> 00:54:41.960
But, going back to the lifetime of this
machine, we started doing experiments

00:54:41.970 --> 00:54:47.140
in the fall of 2016, sort of gingerly
testing it at low stress levels.

00:54:47.140 --> 00:54:51.440
And basically, in 2017,
we started testing it in earnest.

00:54:51.440 --> 00:54:57.330
We’ve done 40 days of testing
throughout this two-plus-year period.

00:54:57.330 --> 00:55:02.750
And each day of testing, we might
do maybe five experimental runs

00:55:02.750 --> 00:55:09.680
with five to 10 events each.
The cumulative slip on this fault –

00:55:09.680 --> 00:55:15.360
it’s maybe 100 or 200
millimeters of slip total.

00:55:16.100 --> 00:55:18.940
So most of the results
that I’ve shown is after

00:55:18.940 --> 00:55:22.220
about 100 or 150
millimeters of slip.

00:55:22.220 --> 00:55:25.400
So that’s the sort of
wear we’re getting.

00:55:25.400 --> 00:55:29.710
And these examples and sequences that
I’m showing today, they are examples.

00:55:29.710 --> 00:55:32.480
They’re examples that illustrate
the process very clearly.

00:55:32.480 --> 00:55:38.400
But we see it quite repeatably.
So I’m not just – I’m not just choosing –

00:55:38.400 --> 00:55:41.860
I’m choosing champion data,
but I believe that it is indicative

00:55:41.860 --> 00:55:45.000
of an underlying phenomenon
that’s quite general.

00:55:47.180 --> 00:55:56.020
[Silence]

00:55:56.020 --> 00:55:58.100
- Very nice talk, Greg.
Two questions.

00:55:58.100 --> 00:56:02.520
One – you kind of alluded to it earlier
in the talk – if you had any comments

00:56:02.520 --> 00:56:07.360
about, if you varied the depth of your
fault, whether on your machine or

00:56:07.360 --> 00:56:11.580
our machine, how that might affect
nucleation or rupture processes with

00:56:11.580 --> 00:56:16.490
more fault area to work with.
And just in – just maybe a general

00:56:16.490 --> 00:56:19.960
comment, how does – how did results
from your machine, the big machine,

00:56:19.960 --> 00:56:24.880
vary from your small machine, like our
machine and then the NIED machine?

00:56:27.960 --> 00:56:32.300
- So our fault is 10 times
longer than it is deep.

00:56:32.300 --> 00:56:36.060
And so I like to think of it
as a one-dimensional problem.

00:56:36.060 --> 00:56:38.760
That whatever is happening at the
top surface is pretty much indicative

00:56:38.760 --> 00:56:41.060
of what’s happening through
the thickness of the slab.

00:56:41.060 --> 00:56:44.590
Now, those free surfaces do have
an effect on what’s going on.

00:56:44.590 --> 00:56:49.570
And my guess is that, as you go
from this 1D geometry to 2D geometry

00:56:49.570 --> 00:56:52.400
and make it thicker,
these two-dimensional effects,

00:56:52.400 --> 00:56:55.720
like we saw with the nucleation
process sort of starting at one –

00:56:55.720 --> 00:56:58.560
the bottom of the slab and
migrating up – those become

00:56:58.560 --> 00:57:02.900
much more pronounced. Things will
probably get more complicated.

00:57:07.080 --> 00:57:11.230
How these results compare to
smaller scale – so I have this –

00:57:11.230 --> 00:57:15.710
this is a tabletop apparatus that I built.
Don’t ever try and build one of

00:57:15.710 --> 00:57:19.320
these machines from scratch.
Always build a small-scale version.

00:57:19.320 --> 00:57:22.810
This was the smartest thing I ever did.
Because I started to realize what could

00:57:22.810 --> 00:57:30.000
go wrong at the small scale before
I really messed up at the big scale.

00:57:30.000 --> 00:57:31.880
On the small scale,
the nucleation length size

00:57:31.880 --> 00:57:37.060
is about the same size as the sample.
So, in this case, the behavior of fast

00:57:37.060 --> 00:57:42.520
and slow occurs not just on confined
events but the entire interface can slip

00:57:42.520 --> 00:57:46.520
in a stick-slip fashion, fast and slow.
And we see the same results, basically.

00:57:46.520 --> 00:57:49.850
So if you kick the small
sample, it can go dynamic.

00:57:49.850 --> 00:57:53.460
And if you loaded it more slowly,
it will slip slowly and

00:57:53.460 --> 00:57:59.200
radiate tremor-like signals.
So, in that sense, it’s very similar.

00:57:59.200 --> 00:58:04.220
If we compare this to the 2-meter block
in Building 4, you know, pretty much

00:58:04.220 --> 00:58:08.160
everything that we see here is not that
different from what we see there.

00:58:08.160 --> 00:58:12.860
We see roughly meter-size
nucleation length scales.

00:58:12.860 --> 00:58:15.371
The only difference is that,
because we’re loading on one end,

00:58:15.371 --> 00:58:18.630
and we’re generating this really
heterogeneous distribution of stress,

00:58:18.630 --> 00:58:23.060
we’re able to see a lot of variation
that wasn’t really possible

00:58:23.060 --> 00:58:25.740
with a 2-meter block.
And we’re sort of putting it

00:58:25.740 --> 00:58:30.430
through a few more tricks
than we did at the 2-meter block.

00:58:30.430 --> 00:58:35.220
But we’re also loading this thing by
hand. So we don’t have quite the smooth

00:58:35.220 --> 00:58:39.740
servo control and the really slow
loading capabilities that we had here.

00:58:39.740 --> 00:58:43.550
So it’s – I think this has been a –
you know, some people will ask about

00:58:43.550 --> 00:58:46.030
repeatability of these experiments.
Like, how repeatable are they?

00:58:46.030 --> 00:58:50.480
Well, if you really wanted to see how
repeatable an experiment would be,

00:58:50.480 --> 00:58:53.670
you would go somewhere else and
try and build another machine that’s

00:58:53.670 --> 00:58:56.380
kind of similar but different
and do the same thing.

00:58:56.380 --> 00:58:59.010
And if you saw the same behavior, then
you would see that it’s quite repeatable,

00:58:59.010 --> 00:59:02.480
and that’s kind of what I did.
And I see a lot of the same behavior.

00:59:02.480 --> 00:59:06.840
A lot of the trends that
I see are very much consistent.

00:59:06.840 --> 00:59:09.280
This thing that Art is saying with
the inverse proportionality

00:59:09.280 --> 00:59:12.500
between normal stress and
nucleation length, we see that.

00:59:12.500 --> 00:59:16.390
So that’s great. It’s reinforcing some
of these views that I’ve sort of

00:59:16.390 --> 00:59:21.610
developed over the past 10 years or so.
But we’re also getting a glimpse at some

00:59:21.610 --> 00:59:26.230
of the complexity that occurs when
you have a larger fault and a more

00:59:26.230 --> 00:59:30.360
complicated stress distribution.
And that glimpse of complexity is

00:59:30.360 --> 00:59:33.920
actually causing me to think maybe –
maybe this nucleation length scale is

00:59:33.920 --> 00:59:38.080
not even the right way of thinking
about how earthquakes initiate.

00:59:40.340 --> 00:59:44.040
[Silence]

00:59:44.040 --> 00:59:46.120
- Do we have any
more questions?

00:59:47.780 --> 00:59:51.800
[Silence]

00:59:51.800 --> 00:59:54.900
All right. Nick, do you want to say
anything about lunch or the schedule?

00:59:54.900 --> 00:59:56.920
- Yeah. We’re going to go to lunch.
We haven’t decided where yet.

00:59:56.920 --> 01:00:00.740
But everyone’s welcome to join us.
And if there’s time afterwards and

01:00:00.740 --> 01:00:04.360
any enthusiasm and if Greg has time,
we might grab a beer someplace too.

01:00:04.360 --> 01:00:08.480
So everybody’s welcome
to that escapade as well.

01:00:10.820 --> 01:00:14.000
- All right. Well, let’s give
another round of applause for Greg.

01:00:14.000 --> 01:00:19.460
[Applause]