WEBVTT

NOTE Narration transcript. Text is the script synthesized by minimax (speech-2.8-hd, voice English_expressive_narrator); each cue was synthesized as its own audio file and is timed by that file's measured duration. Verified at 105.0% of expected length, 138.1 wpm (median cue 140.6 wpm). Script source: public/videos/critical-minerals-pfas-and-the-remediation-imperative.vtt at commit 4641d96 (pre-overwrite narration prose)

1
00:00:00.000 --> 00:00:02.940
The energy transition needs a lot more critical minerals.

2
00:00:03.240 --> 00:00:06.390
Our water needs a lot fewer forever chemicals.

3
00:00:06.690 --> 00:00:08.520
Totally different crises, right?

4
00:00:08.820 --> 00:00:16.580
At the atomic level, they are weirdly the same problem: getting a material to grab, release, or break exactly the right thing.

5
00:00:16.880 --> 00:00:22.000
Clean-energy mineral demand could grow four to six times by 2040.

6
00:00:22.300 --> 00:00:26.050
Mining alone cannot get us there cleanly or quickly.

7
00:00:26.350 --> 00:00:31.760
So recycling, urban mining, and direct lithium extraction have to work.

8
00:00:32.060 --> 00:00:39.650
But that requires sorbents and extractants precise enough to separate nearly identical ions in messy real-world streams.

9
00:00:39.950 --> 00:00:42.250
PFAS flips the challenge around.

10
00:00:42.550 --> 00:00:49.590
First, capture PFOA and PFOS down to the EPA limit: four nanograms per liter each.

11
00:00:49.890 --> 00:00:52.240
Then actually destroy them.

12
00:00:52.540 --> 00:00:59.780
That second step is brutal because the carbon–fluorine bond packs about four hundred eighty-five kilojoules per mole.

13
00:01:00.080 --> 00:01:04.040
Today’s filters mostly move the problem into contaminated media.

14
00:01:04.340 --> 00:01:11.840
We could screen millions of recovery materials and catalysts, except the fast models tend to soften the exact physics that matters.

15
00:01:12.140 --> 00:01:24.300
At pores, vacancies, surfaces, and transition states, raw universal machine-learning potentials can misrepresent the energy landscape by fifteen to sixty percent.

16
00:01:24.600 --> 00:01:27.870
Great-looking candidates become expensive false positives.

17
00:01:28.170 --> 00:01:34.780
The correction measures that error against local atomic coordination, then corrects energies and forces at runtime.

18
00:01:35.080 --> 00:01:48.860
Now the same fast screen can rank lithium-selective pores, cobalt-versus-nickel extractants, PFAS sorbents, and defluorination catalysts without pretending every atomic neighborhood behaves like a perfect bulk crystal.

19
00:01:49.160 --> 00:01:55.590
On the mineral side, corrected binding and migration energies guide separation and direct battery recycling.

20
00:01:55.890 --> 00:02:03.230
On the PFAS side, corrected barriers expose catalysts that will fail or poison themselves with fluoride.

21
00:02:03.530 --> 00:02:10.770
Machine-checked boundaries also flag predictions the data cannot support, before anybody builds a costly experiment around them.

22
00:02:11.070 --> 00:02:14.480
One campaign puts scarce atoms back into circulation.

23
00:02:14.780 --> 00:02:17.000
The other takes harmful ones out.

24
00:02:17.300 --> 00:02:21.310
Both move faster when we correct the same hidden error in the models.

25
00:02:21.610 --> 00:02:23.100
Recover what we need.

26
00:02:23.400 --> 00:02:25.060
Destroy what we don’t.

27
00:02:25.360 --> 00:02:27.689
Read the full case in the article.